Skip to content

MusicLDM

MusicLDM was proposed in MusicLDM: Enhancing Novelty in Text-to-Music Generation Using Beat-Synchronous Mixup Strategies by Ke Chen, Yusong Wu, Haohe Liu, Marianna Nezhurina, Taylor Berg-Kirkpatrick, Shlomo Dubnov. MusicLDM takes a text prompt as input and predicts the corresponding music sample.

Inspired by Stable Diffusion and AudioLDM, MusicLDM is a text-to-music latent diffusion model (LDM) that learns continuous audio representations from CLAP latents.

MusicLDM is trained on a corpus of 466 hours of music data. Beat-synchronous data augmentation strategies are applied to the music samples, both in the time domain and in the latent space. Using beat-synchronous data augmentation strategies encourages the model to interpolate between the training samples, but stay within the domain of the training data. The result is generated music that is more diverse while staying faithful to the corresponding style.

The abstract of the paper is the following:

Diffusion models have shown promising results in cross-modal generation tasks, including text-to-image and text-to-audio generation. However, generating music, as a special type of audio, presents unique challenges due to limited availability of music data and sensitive issues related to copyright and plagiarism. In this paper, to tackle these challenges, we first construct a state-of-the-art text-to-music model, MusicLDM, that adapts Stable Diffusion and AudioLDM architectures to the music domain. We achieve this by retraining the contrastive language-audio pretraining model (CLAP) and the Hifi-GAN vocoder, as components of MusicLDM, on a collection of music data samples. Then, to address the limitations of training data and to avoid plagiarism, we leverage a beat tracking model and propose two different mixup strategies for data augmentation: beat-synchronous audio mixup and beat-synchronous latent mixup, which recombine training audio directly or via a latent embeddings space, respectively. Such mixup strategies encourage the model to interpolate between musical training samples and generate new music within the convex hull of the training data, making the generated music more diverse while still staying faithful to the corresponding style. In addition to popular evaluation metrics, we design several new evaluation metrics based on CLAP score to demonstrate that our proposed MusicLDM and beat-synchronous mixup strategies improve both the quality and novelty of generated music, as well as the correspondence between input text and generated music.

This pipeline was contributed by sanchit-gandhi.

Tips

When constructing a prompt, keep in mind:

  • Descriptive prompt inputs work best; use adjectives to describe the sound (for example, "high quality" or "clear") and make the prompt context specific where possible (e.g. "melodic techno with a fast beat and synths" works better than "techno").
  • Using a negative prompt can significantly improve the quality of the generated audio. Try using a negative prompt of "low quality, average quality".

During inference:

  • The quality of the generated audio sample can be controlled by the num_inference_steps argument; higher steps give higher quality audio at the expense of slower inference.
  • Multiple waveforms can be generated in one go: set num_waveforms_per_prompt to a value greater than 1 to enable. Automatic scoring will be performed between the generated waveforms and prompt text, and the audios ranked from best to worst accordingly.
  • The length of the generated audio sample can be controlled by varying the audio_length_in_s argument.

Tip

Make sure to check out the Schedulers guide to learn how to explore the tradeoff between scheduler speed and quality, and see the reuse components across pipelines section to learn how to efficiently load the same components into multiple pipelines.

mindone.diffusers.MusicLDMPipeline

Bases: DiffusionPipeline, StableDiffusionMixin

Pipeline for text-to-audio generation using MusicLDM.

This model inherits from [DiffusionPipeline]. Check the superclass documentation for the generic methods implemented for all pipelines (downloading, saving, running on a particular etc.).

PARAMETER DESCRIPTION
vae

Variational Auto-Encoder (VAE) model to encode and decode images to and from latent representations.

TYPE: [`AutoencoderKL`]

text_encoder

Frozen text-audio embedding model (ClapTextModel), specifically the laion/clap-htsat-unfused variant.

TYPE: [`~transformers.ClapModel`]

tokenizer

A [~transformers.RobertaTokenizer] to tokenize text.

TYPE: [`PreTrainedTokenizer`]

feature_extractor

Feature extractor to compute mel-spectrograms from audio waveforms.

TYPE: [`~transformers.ClapFeatureExtractor`]

unet

A UNet2DConditionModel to denoise the encoded audio latents.

TYPE: [`UNet2DConditionModel`]

scheduler

A scheduler to be used in combination with unet to denoise the encoded audio latents. Can be one of [DDIMScheduler], [LMSDiscreteScheduler], or [PNDMScheduler].

TYPE: [`SchedulerMixin`]

vocoder

Vocoder of class SpeechT5HifiGan.

TYPE: [`~transformers.SpeechT5HifiGan`]

Source code in mindone/diffusers/pipelines/musicldm/pipeline_musicldm.py
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
class MusicLDMPipeline(DiffusionPipeline, StableDiffusionMixin):
    r"""
    Pipeline for text-to-audio generation using MusicLDM.

    This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods
    implemented for all pipelines (downloading, saving, running on a particular etc.).

    Args:
        vae ([`AutoencoderKL`]):
            Variational Auto-Encoder (VAE) model to encode and decode images to and from latent representations.
        text_encoder ([`~transformers.ClapModel`]):
            Frozen text-audio embedding model (`ClapTextModel`), specifically the
            [laion/clap-htsat-unfused](https://huggingface.co/laion/clap-htsat-unfused) variant.
        tokenizer ([`PreTrainedTokenizer`]):
            A [`~transformers.RobertaTokenizer`] to tokenize text.
        feature_extractor ([`~transformers.ClapFeatureExtractor`]):
            Feature extractor to compute mel-spectrograms from audio waveforms.
        unet ([`UNet2DConditionModel`]):
            A `UNet2DConditionModel` to denoise the encoded audio latents.
        scheduler ([`SchedulerMixin`]):
            A scheduler to be used in combination with `unet` to denoise the encoded audio latents. Can be one of
            [`DDIMScheduler`], [`LMSDiscreteScheduler`], or [`PNDMScheduler`].
        vocoder ([`~transformers.SpeechT5HifiGan`]):
            Vocoder of class `SpeechT5HifiGan`.
    """

    def __init__(
        self,
        vae: AutoencoderKL,
        text_encoder: Union[ClapTextModelWithProjection, ClapModel],
        tokenizer: Union[RobertaTokenizer, RobertaTokenizerFast],
        feature_extractor: Optional[ClapFeatureExtractor],
        unet: UNet2DConditionModel,
        scheduler: KarrasDiffusionSchedulers,
        vocoder: SpeechT5HifiGan,
    ):
        super().__init__()

        self.register_modules(
            vae=vae,
            text_encoder=text_encoder,
            tokenizer=tokenizer,
            feature_extractor=feature_extractor,
            unet=unet,
            scheduler=scheduler,
            vocoder=vocoder,
        )
        self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1)

    def _encode_prompt(
        self,
        prompt,
        num_waveforms_per_prompt,
        do_classifier_free_guidance,
        negative_prompt=None,
        prompt_embeds: Optional[ms.Tensor] = None,
        negative_prompt_embeds: Optional[ms.Tensor] = None,
    ):
        r"""
        Encodes the prompt into text encoder hidden states.

        Args:
            prompt (`str` or `List[str]`, *optional*):
                prompt to be encoded
            num_waveforms_per_prompt (`int`):
                number of waveforms that should be generated per prompt
            do_classifier_free_guidance (`bool`):
                whether to use classifier free guidance or not
            negative_prompt (`str` or `List[str]`, *optional*):
                The prompt or prompts not to guide the audio generation. If not defined, one has to pass
                `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
                less than `1`).
            prompt_embeds (`ms.Tensor`, *optional*):
                Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
                provided, text embeddings will be generated from `prompt` input argument.
            negative_prompt_embeds (`ms.Tensor`, *optional*):
                Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
                weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
                argument.
        """
        if prompt is not None and isinstance(prompt, str):
            batch_size = 1
        elif prompt is not None and isinstance(prompt, list):
            batch_size = len(prompt)
        else:
            batch_size = prompt_embeds.shape[0]

        if prompt_embeds is None:
            text_inputs = self.tokenizer(
                prompt,
                padding="max_length",
                max_length=self.tokenizer.model_max_length,
                truncation=True,
                return_tensors="np",
            )
            text_input_ids = ms.Tensor(text_inputs.input_ids)
            attention_mask = ms.Tensor(text_inputs.attention_mask)
            untruncated_ids = ms.Tensor(self.tokenizer(prompt, padding="longest", return_tensors="np").input_ids)

            if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not ops.equal(text_input_ids, untruncated_ids):
                removed_text = self.tokenizer.batch_decode(untruncated_ids[:, self.tokenizer.model_max_length - 1 : -1])
                logger.warning(
                    "The following part of your input was truncated because CLAP can only handle sequences up to"
                    f" {self.tokenizer.model_max_length} tokens: {removed_text}"
                )

            prompt_embeds = self.text_encoder.get_text_features(
                text_input_ids,
                attention_mask=attention_mask,
            )

        prompt_embeds = prompt_embeds.to(dtype=self.text_encoder.text_model.dtype)

        (
            bs_embed,
            seq_len,
        ) = prompt_embeds.shape
        # duplicate text embeddings for each generation per prompt, using mps friendly method
        prompt_embeds = prompt_embeds.repeat(1, num_waveforms_per_prompt)
        prompt_embeds = prompt_embeds.view(bs_embed * num_waveforms_per_prompt, seq_len)

        # get unconditional embeddings for classifier free guidance
        if do_classifier_free_guidance and negative_prompt_embeds is None:
            uncond_tokens: List[str]
            if negative_prompt is None:
                uncond_tokens = [""] * batch_size
            elif type(prompt) is not type(negative_prompt):
                raise TypeError(
                    f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
                    f" {type(prompt)}."
                )
            elif isinstance(negative_prompt, str):
                uncond_tokens = [negative_prompt]
            elif batch_size != len(negative_prompt):
                raise ValueError(
                    f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:"
                    f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches"
                    " the batch size of `prompt`."
                )
            else:
                uncond_tokens = negative_prompt

            max_length = prompt_embeds.shape[1]
            uncond_input = self.tokenizer(
                uncond_tokens,
                padding="max_length",
                max_length=max_length,
                truncation=True,
                return_tensors="np",
            )

            uncond_input_ids = ms.Tensor(uncond_input.input_ids)
            attention_mask = ms.Tensor(uncond_input.attention_mask)

            negative_prompt_embeds = self.text_encoder.get_text_features(
                uncond_input_ids,
                attention_mask=attention_mask,
            )

        if do_classifier_free_guidance:
            # duplicate unconditional embeddings for each generation per prompt, using mps friendly method
            seq_len = negative_prompt_embeds.shape[1]

            negative_prompt_embeds = negative_prompt_embeds.to(dtype=self.text_encoder.text_model.dtype)

            negative_prompt_embeds = negative_prompt_embeds.repeat(1, num_waveforms_per_prompt)
            negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_waveforms_per_prompt, seq_len)

            # For classifier free guidance, we need to do two forward passes.
            # Here we concatenate the unconditional and text embeddings into a single batch
            # to avoid doing two forward passes
            prompt_embeds = ops.cat([negative_prompt_embeds, prompt_embeds])

        return prompt_embeds

    # Copied from diffusers.pipelines.audioldm.pipeline_audioldm.AudioLDMPipeline.mel_spectrogram_to_waveform
    def mel_spectrogram_to_waveform(self, mel_spectrogram):
        if mel_spectrogram.dim() == 4:
            mel_spectrogram = mel_spectrogram.squeeze(1)

        waveform = self.vocoder(mel_spectrogram)
        # we always cast to float32 as this does not cause significant overhead and is compatible with bfloat16
        waveform = waveform.float()
        return waveform

    # Copied from diffusers.pipelines.audioldm2.pipeline_audioldm2.AudioLDM2Pipeline.score_waveforms
    def score_waveforms(self, text, audio, num_waveforms_per_prompt, dtype):
        # if not is_librosa_available():
        #     logger.info(
        #         "Automatic scoring of the generated audio waveforms against the input prompt text requires the "
        #         "`librosa` package to resample the generated waveforms. Returning the audios in the order they were "
        #         "generated. To enable automatic scoring, install `librosa` with: `pip install librosa`."
        #     )
        #     return audio
        inputs = ms.Tensor(self.tokenizer(text, return_tensors="np", padding=True))
        resampled_audio = librosa.resample(
            audio.numpy(), orig_sr=self.vocoder.config.sampling_rate, target_sr=self.feature_extractor.sampling_rate
        )
        inputs["input_features"] = ms.Tensor(
            self.feature_extractor(
                list(resampled_audio), return_tensors="np", sampling_rate=self.feature_extractor.sampling_rate
            ).input_features.type(dtype)
        )
        inputs = inputs

        # compute the audio-text similarity score using the CLAP model
        logits_per_text = self.text_encoder(**inputs).logits_per_text
        # sort by the highest matching generations per prompt
        indices = ops.argsort(logits_per_text, dim=1, descending=True)[:, :num_waveforms_per_prompt]
        audio = ops.index_select(audio, 0, indices.reshape(-1))
        return audio

    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_extra_step_kwargs
    def prepare_extra_step_kwargs(self, generator, eta):
        # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
        # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
        # eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502
        # and should be between [0, 1]

        accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
        extra_step_kwargs = {}
        if accepts_eta:
            extra_step_kwargs["eta"] = eta

        # check if the scheduler accepts generator
        accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys())
        if accepts_generator:
            extra_step_kwargs["generator"] = generator
        return extra_step_kwargs

    # Copied from diffusers.pipelines.audioldm.pipeline_audioldm.AudioLDMPipeline.check_inputs
    def check_inputs(
        self,
        prompt,
        audio_length_in_s,
        vocoder_upsample_factor,
        callback_steps,
        negative_prompt=None,
        prompt_embeds=None,
        negative_prompt_embeds=None,
    ):
        min_audio_length_in_s = vocoder_upsample_factor * self.vae_scale_factor
        if audio_length_in_s < min_audio_length_in_s:
            raise ValueError(
                f"`audio_length_in_s` has to be a positive value greater than or equal to {min_audio_length_in_s}, but "
                f"is {audio_length_in_s}."
            )

        if self.vocoder.config.model_in_dim % self.vae_scale_factor != 0:
            raise ValueError(
                f"The number of frequency bins in the vocoder's log-mel spectrogram has to be divisible by the "
                f"VAE scale factor, but got {self.vocoder.config.model_in_dim} bins and a scale factor of "
                f"{self.vae_scale_factor}."
            )

        if (callback_steps is None) or (
            callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0)
        ):
            raise ValueError(
                f"`callback_steps` has to be a positive integer but is {callback_steps} of type"
                f" {type(callback_steps)}."
            )

        if prompt is not None and prompt_embeds is not None:
            raise ValueError(
                f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
                " only forward one of the two."
            )
        elif prompt is None and prompt_embeds is None:
            raise ValueError(
                "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined."
            )
        elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)):
            raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")

        if negative_prompt is not None and negative_prompt_embeds is not None:
            raise ValueError(
                f"Cannot forward both `negative_prompt`: {negative_prompt} and `negative_prompt_embeds`:"
                f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
            )

        if prompt_embeds is not None and negative_prompt_embeds is not None:
            if prompt_embeds.shape != negative_prompt_embeds.shape:
                raise ValueError(
                    "`prompt_embeds` and `negative_prompt_embeds` must have the same shape when passed directly, but"
                    f" got: `prompt_embeds` {prompt_embeds.shape} != `negative_prompt_embeds`"
                    f" {negative_prompt_embeds.shape}."
                )

    # Copied from diffusers.pipelines.audioldm.pipeline_audioldm.AudioLDMPipeline.prepare_latents
    def prepare_latents(self, batch_size, num_channels_latents, height, dtype, generator, latents=None):
        shape = (
            batch_size,
            num_channels_latents,
            int(height) // self.vae_scale_factor,
            int(self.vocoder.config.model_in_dim) // self.vae_scale_factor,
        )
        if isinstance(generator, list) and len(generator) != batch_size:
            raise ValueError(
                f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
                f" size of {batch_size}. Make sure the batch size matches the length of the generators."
            )

        if latents is None:
            latents = randn_tensor(shape, generator=generator, dtype=dtype)
        else:
            latents = latents

        # scale the initial noise by the standard deviation required by the scheduler
        latents = latents * self.scheduler.init_noise_sigma
        return latents

    def __call__(
        self,
        prompt: Union[str, List[str]] = None,
        audio_length_in_s: Optional[float] = None,
        num_inference_steps: int = 200,
        guidance_scale: float = 2.0,
        negative_prompt: Optional[Union[str, List[str]]] = None,
        num_waveforms_per_prompt: Optional[int] = 1,
        eta: float = 0.0,
        generator: Optional[Union[np.random.Generator, List[np.random.Generator]]] = None,
        latents: Optional[ms.Tensor] = None,
        prompt_embeds: Optional[ms.Tensor] = None,
        negative_prompt_embeds: Optional[ms.Tensor] = None,
        return_dict: bool = True,
        callback: Optional[Callable[[int, int, ms.Tensor], None]] = None,
        callback_steps: Optional[int] = 1,
        cross_attention_kwargs: Optional[Dict[str, Any]] = None,
        output_type: Optional[str] = "np",
    ):
        r"""
        The call function to the pipeline for generation.

        Args:
            prompt (`str` or `List[str]`, *optional*):
                The prompt or prompts to guide audio generation. If not defined, you need to pass `prompt_embeds`.
            audio_length_in_s (`int`, *optional*, defaults to 10.24):
                The length of the generated audio sample in seconds.
            num_inference_steps (`int`, *optional*, defaults to 200):
                The number of denoising steps. More denoising steps usually lead to a higher quality audio at the
                expense of slower inference.
            guidance_scale (`float`, *optional*, defaults to 2.0):
                A higher guidance scale value encourages the model to generate audio that is closely linked to the text
                `prompt` at the expense of lower sound quality. Guidance scale is enabled when `guidance_scale > 1`.
            negative_prompt (`str` or `List[str]`, *optional*):
                The prompt or prompts to guide what to not include in audio generation. If not defined, you need to
                pass `negative_prompt_embeds` instead. Ignored when not using guidance (`guidance_scale < 1`).
            num_waveforms_per_prompt (`int`, *optional*, defaults to 1):
                The number of waveforms to generate per prompt. If `num_waveforms_per_prompt > 1`, the text encoding
                model is a joint text-audio model ([`~transformers.ClapModel`]), and the tokenizer is a
                `[~transformers.ClapProcessor]`, then automatic scoring will be performed between the generated outputs
                and the input text. This scoring ranks the generated waveforms based on their cosine similarity to text
                input in the joint text-audio embedding space.
            eta (`float`, *optional*, defaults to 0.0):
                Corresponds to parameter eta (η) from the [DDIM](https://arxiv.org/abs/2010.02502) paper. Only applies
                to the [`~schedulers.DDIMScheduler`], and is ignored in other schedulers.
            generator (`np.random.Generator` or `List[np.random.Generator]`, *optional*):
                A [`np.random.Generator`](https://pytorch.org/docs/stable/generated/np.random.Generator.html) to make
                generation deterministic.
            latents (`ms.Tensor`, *optional*):
                Pre-generated noisy latents sampled from a Gaussian distribution, to be used as inputs for image
                generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
                tensor is generated by sampling using the supplied random `generator`.
            prompt_embeds (`ms.Tensor`, *optional*):
                Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not
                provided, text embeddings are generated from the `prompt` input argument.
            negative_prompt_embeds (`ms.Tensor`, *optional*):
                Pre-generated negative text embeddings. Can be used to easily tweak text inputs (prompt weighting). If
                not provided, `negative_prompt_embeds` are generated from the `negative_prompt` input argument.
            return_dict (`bool`, *optional*, defaults to `True`):
                Whether or not to return a [`~pipelines.AudioPipelineOutput`] instead of a plain tuple.
            callback (`Callable`, *optional*):
                A function that calls every `callback_steps` steps during inference. The function is called with the
                following arguments: `callback(step: int, timestep: int, latents: ms.Tensor)`.
            callback_steps (`int`, *optional*, defaults to 1):
                The frequency at which the `callback` function is called. If not specified, the callback is called at
                every step.
            cross_attention_kwargs (`dict`, *optional*):
                A kwargs dictionary that if specified is passed along to the [`AttentionProcessor`] as defined in
                [`self.processor`](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
            output_type (`str`, *optional*, defaults to `"np"`):
                The output format of the generated audio. Choose between `"np"` to return a NumPy `np.ndarray` or
                `"np"` to return a PyTorch `ms.Tensor` object. Set to `"latent"` to return the latent diffusion
                model (LDM) output.

        Examples:

        Returns:
            [`~pipelines.AudioPipelineOutput`] or `tuple`:
                If `return_dict` is `True`, [`~pipelines.AudioPipelineOutput`] is returned, otherwise a `tuple` is
                returned where the first element is a list with the generated audio.
        """
        # 0. Convert audio input length from seconds to spectrogram height
        vocoder_upsample_factor = np.prod(self.vocoder.config.upsample_rates) / self.vocoder.config.sampling_rate

        if audio_length_in_s is None:
            audio_length_in_s = self.unet.config.sample_size * self.vae_scale_factor * vocoder_upsample_factor

        height = int(audio_length_in_s / vocoder_upsample_factor)

        original_waveform_length = int(audio_length_in_s * self.vocoder.config.sampling_rate)
        if height % self.vae_scale_factor != 0:
            height = int(np.ceil(height / self.vae_scale_factor)) * self.vae_scale_factor
            logger.info(
                f"Audio length in seconds {audio_length_in_s} is increased to {height * vocoder_upsample_factor} "
                f"so that it can be handled by the model. It will be cut to {audio_length_in_s} after the "
                f"denoising process."
            )

        # 1. Check inputs. Raise error if not correct
        self.check_inputs(
            prompt,
            audio_length_in_s,
            vocoder_upsample_factor,
            callback_steps,
            negative_prompt,
            prompt_embeds,
            negative_prompt_embeds,
        )

        # 2. Define call parameters
        if prompt is not None and isinstance(prompt, str):
            batch_size = 1
        elif prompt is not None and isinstance(prompt, list):
            batch_size = len(prompt)
        else:
            batch_size = prompt_embeds.shape[0]

        # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
        # of the Imagen paper: https://arxiv.org/pdf/2205.11487.pdf . `guidance_scale = 1`
        # corresponds to doing no classifier free guidance.
        do_classifier_free_guidance = guidance_scale > 1.0

        # 3. Encode input prompt
        prompt_embeds = self._encode_prompt(
            prompt,
            num_waveforms_per_prompt,
            do_classifier_free_guidance,
            negative_prompt,
            prompt_embeds=prompt_embeds,
            negative_prompt_embeds=negative_prompt_embeds,
        )

        # 4. Prepare timesteps
        self.scheduler.set_timesteps(num_inference_steps)
        timesteps = self.scheduler.timesteps

        # 5. Prepare latent variables
        num_channels_latents = self.unet.config.in_channels
        latents = self.prepare_latents(
            batch_size * num_waveforms_per_prompt,
            num_channels_latents,
            height,
            prompt_embeds.dtype,
            generator,
            latents,
        )

        # 6. Prepare extra step kwargs
        extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)

        # 7. Denoising loop
        num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order
        with self.progress_bar(total=num_inference_steps) as progress_bar:
            for i, t in enumerate(timesteps):
                # expand the latents if we are doing classifier free guidance
                latent_model_input = ops.cat([latents] * 2) if do_classifier_free_guidance else latents
                latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)

                # predict the noise residual
                noise_pred = self.unet(
                    latent_model_input,
                    t,
                    encoder_hidden_states=None,
                    class_labels=prompt_embeds,
                    cross_attention_kwargs=cross_attention_kwargs,
                    return_dict=False,
                )[0]

                # perform guidance
                if do_classifier_free_guidance:
                    noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
                    noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)

                # compute the previous noisy sample x_t -> x_t-1
                latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs)[0]

                # call the callback, if provided
                if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
                    progress_bar.update()
                    if callback is not None and i % callback_steps == 0:
                        step_idx = i // getattr(self.scheduler, "order", 1)
                        callback(step_idx, t, latents)

        # 8. Post-processing
        if not output_type == "latent":
            latents = 1 / self.vae.config.scaling_factor * latents
            mel_spectrogram = self.vae.decode(latents)[0]
        else:
            return AudioPipelineOutput(audios=latents)

        audio = self.mel_spectrogram_to_waveform(mel_spectrogram)

        audio = audio[:, :original_waveform_length]

        # 9. Automatic scoring
        if num_waveforms_per_prompt > 1 and prompt is not None:
            audio = self.score_waveforms(
                text=prompt,
                audio=audio,
                num_waveforms_per_prompt=num_waveforms_per_prompt,
                dtype=prompt_embeds.dtype,
            )

        if output_type == "np":
            audio = audio.numpy()

        if not return_dict:
            return (audio,)

        return AudioPipelineOutput(audios=audio)

mindone.diffusers.MusicLDMPipeline.__call__(prompt=None, audio_length_in_s=None, num_inference_steps=200, guidance_scale=2.0, negative_prompt=None, num_waveforms_per_prompt=1, eta=0.0, generator=None, latents=None, prompt_embeds=None, negative_prompt_embeds=None, return_dict=True, callback=None, callback_steps=1, cross_attention_kwargs=None, output_type='np')

The call function to the pipeline for generation.

PARAMETER DESCRIPTION
prompt

The prompt or prompts to guide audio generation. If not defined, you need to pass prompt_embeds.

TYPE: `str` or `List[str]`, *optional* DEFAULT: None

audio_length_in_s

The length of the generated audio sample in seconds.

TYPE: `int`, *optional*, defaults to 10.24 DEFAULT: None

num_inference_steps

The number of denoising steps. More denoising steps usually lead to a higher quality audio at the expense of slower inference.

TYPE: `int`, *optional*, defaults to 200 DEFAULT: 200

guidance_scale

A higher guidance scale value encourages the model to generate audio that is closely linked to the text prompt at the expense of lower sound quality. Guidance scale is enabled when guidance_scale > 1.

TYPE: `float`, *optional*, defaults to 2.0 DEFAULT: 2.0

negative_prompt

The prompt or prompts to guide what to not include in audio generation. If not defined, you need to pass negative_prompt_embeds instead. Ignored when not using guidance (guidance_scale < 1).

TYPE: `str` or `List[str]`, *optional* DEFAULT: None

num_waveforms_per_prompt

The number of waveforms to generate per prompt. If num_waveforms_per_prompt > 1, the text encoding model is a joint text-audio model ([~transformers.ClapModel]), and the tokenizer is a [~transformers.ClapProcessor], then automatic scoring will be performed between the generated outputs and the input text. This scoring ranks the generated waveforms based on their cosine similarity to text input in the joint text-audio embedding space.

TYPE: `int`, *optional*, defaults to 1 DEFAULT: 1

eta

Corresponds to parameter eta (η) from the DDIM paper. Only applies to the [~schedulers.DDIMScheduler], and is ignored in other schedulers.

TYPE: `float`, *optional*, defaults to 0.0 DEFAULT: 0.0

generator

A np.random.Generator to make generation deterministic.

TYPE: `np.random.Generator` or `List[np.random.Generator]`, *optional* DEFAULT: None

latents

Pre-generated noisy latents sampled from a Gaussian distribution, to be used as inputs for image generation. Can be used to tweak the same generation with different prompts. If not provided, a latents tensor is generated by sampling using the supplied random generator.

TYPE: `ms.Tensor`, *optional* DEFAULT: None

prompt_embeds

Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not provided, text embeddings are generated from the prompt input argument.

TYPE: `ms.Tensor`, *optional* DEFAULT: None

negative_prompt_embeds

Pre-generated negative text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not provided, negative_prompt_embeds are generated from the negative_prompt input argument.

TYPE: `ms.Tensor`, *optional* DEFAULT: None

return_dict

Whether or not to return a [~pipelines.AudioPipelineOutput] instead of a plain tuple.

TYPE: `bool`, *optional*, defaults to `True` DEFAULT: True

callback

A function that calls every callback_steps steps during inference. The function is called with the following arguments: callback(step: int, timestep: int, latents: ms.Tensor).

TYPE: `Callable`, *optional* DEFAULT: None

callback_steps

The frequency at which the callback function is called. If not specified, the callback is called at every step.

TYPE: `int`, *optional*, defaults to 1 DEFAULT: 1

cross_attention_kwargs

A kwargs dictionary that if specified is passed along to the [AttentionProcessor] as defined in self.processor.

TYPE: `dict`, *optional* DEFAULT: None

output_type

The output format of the generated audio. Choose between "np" to return a NumPy np.ndarray or "np" to return a PyTorch ms.Tensor object. Set to "latent" to return the latent diffusion model (LDM) output.

TYPE: `str`, *optional*, defaults to `"np"` DEFAULT: 'np'

RETURNS DESCRIPTION

[~pipelines.AudioPipelineOutput] or tuple: If return_dict is True, [~pipelines.AudioPipelineOutput] is returned, otherwise a tuple is returned where the first element is a list with the generated audio.

Source code in mindone/diffusers/pipelines/musicldm/pipeline_musicldm.py
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
def __call__(
    self,
    prompt: Union[str, List[str]] = None,
    audio_length_in_s: Optional[float] = None,
    num_inference_steps: int = 200,
    guidance_scale: float = 2.0,
    negative_prompt: Optional[Union[str, List[str]]] = None,
    num_waveforms_per_prompt: Optional[int] = 1,
    eta: float = 0.0,
    generator: Optional[Union[np.random.Generator, List[np.random.Generator]]] = None,
    latents: Optional[ms.Tensor] = None,
    prompt_embeds: Optional[ms.Tensor] = None,
    negative_prompt_embeds: Optional[ms.Tensor] = None,
    return_dict: bool = True,
    callback: Optional[Callable[[int, int, ms.Tensor], None]] = None,
    callback_steps: Optional[int] = 1,
    cross_attention_kwargs: Optional[Dict[str, Any]] = None,
    output_type: Optional[str] = "np",
):
    r"""
    The call function to the pipeline for generation.

    Args:
        prompt (`str` or `List[str]`, *optional*):
            The prompt or prompts to guide audio generation. If not defined, you need to pass `prompt_embeds`.
        audio_length_in_s (`int`, *optional*, defaults to 10.24):
            The length of the generated audio sample in seconds.
        num_inference_steps (`int`, *optional*, defaults to 200):
            The number of denoising steps. More denoising steps usually lead to a higher quality audio at the
            expense of slower inference.
        guidance_scale (`float`, *optional*, defaults to 2.0):
            A higher guidance scale value encourages the model to generate audio that is closely linked to the text
            `prompt` at the expense of lower sound quality. Guidance scale is enabled when `guidance_scale > 1`.
        negative_prompt (`str` or `List[str]`, *optional*):
            The prompt or prompts to guide what to not include in audio generation. If not defined, you need to
            pass `negative_prompt_embeds` instead. Ignored when not using guidance (`guidance_scale < 1`).
        num_waveforms_per_prompt (`int`, *optional*, defaults to 1):
            The number of waveforms to generate per prompt. If `num_waveforms_per_prompt > 1`, the text encoding
            model is a joint text-audio model ([`~transformers.ClapModel`]), and the tokenizer is a
            `[~transformers.ClapProcessor]`, then automatic scoring will be performed between the generated outputs
            and the input text. This scoring ranks the generated waveforms based on their cosine similarity to text
            input in the joint text-audio embedding space.
        eta (`float`, *optional*, defaults to 0.0):
            Corresponds to parameter eta (η) from the [DDIM](https://arxiv.org/abs/2010.02502) paper. Only applies
            to the [`~schedulers.DDIMScheduler`], and is ignored in other schedulers.
        generator (`np.random.Generator` or `List[np.random.Generator]`, *optional*):
            A [`np.random.Generator`](https://pytorch.org/docs/stable/generated/np.random.Generator.html) to make
            generation deterministic.
        latents (`ms.Tensor`, *optional*):
            Pre-generated noisy latents sampled from a Gaussian distribution, to be used as inputs for image
            generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
            tensor is generated by sampling using the supplied random `generator`.
        prompt_embeds (`ms.Tensor`, *optional*):
            Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not
            provided, text embeddings are generated from the `prompt` input argument.
        negative_prompt_embeds (`ms.Tensor`, *optional*):
            Pre-generated negative text embeddings. Can be used to easily tweak text inputs (prompt weighting). If
            not provided, `negative_prompt_embeds` are generated from the `negative_prompt` input argument.
        return_dict (`bool`, *optional*, defaults to `True`):
            Whether or not to return a [`~pipelines.AudioPipelineOutput`] instead of a plain tuple.
        callback (`Callable`, *optional*):
            A function that calls every `callback_steps` steps during inference. The function is called with the
            following arguments: `callback(step: int, timestep: int, latents: ms.Tensor)`.
        callback_steps (`int`, *optional*, defaults to 1):
            The frequency at which the `callback` function is called. If not specified, the callback is called at
            every step.
        cross_attention_kwargs (`dict`, *optional*):
            A kwargs dictionary that if specified is passed along to the [`AttentionProcessor`] as defined in
            [`self.processor`](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
        output_type (`str`, *optional*, defaults to `"np"`):
            The output format of the generated audio. Choose between `"np"` to return a NumPy `np.ndarray` or
            `"np"` to return a PyTorch `ms.Tensor` object. Set to `"latent"` to return the latent diffusion
            model (LDM) output.

    Examples:

    Returns:
        [`~pipelines.AudioPipelineOutput`] or `tuple`:
            If `return_dict` is `True`, [`~pipelines.AudioPipelineOutput`] is returned, otherwise a `tuple` is
            returned where the first element is a list with the generated audio.
    """
    # 0. Convert audio input length from seconds to spectrogram height
    vocoder_upsample_factor = np.prod(self.vocoder.config.upsample_rates) / self.vocoder.config.sampling_rate

    if audio_length_in_s is None:
        audio_length_in_s = self.unet.config.sample_size * self.vae_scale_factor * vocoder_upsample_factor

    height = int(audio_length_in_s / vocoder_upsample_factor)

    original_waveform_length = int(audio_length_in_s * self.vocoder.config.sampling_rate)
    if height % self.vae_scale_factor != 0:
        height = int(np.ceil(height / self.vae_scale_factor)) * self.vae_scale_factor
        logger.info(
            f"Audio length in seconds {audio_length_in_s} is increased to {height * vocoder_upsample_factor} "
            f"so that it can be handled by the model. It will be cut to {audio_length_in_s} after the "
            f"denoising process."
        )

    # 1. Check inputs. Raise error if not correct
    self.check_inputs(
        prompt,
        audio_length_in_s,
        vocoder_upsample_factor,
        callback_steps,
        negative_prompt,
        prompt_embeds,
        negative_prompt_embeds,
    )

    # 2. Define call parameters
    if prompt is not None and isinstance(prompt, str):
        batch_size = 1
    elif prompt is not None and isinstance(prompt, list):
        batch_size = len(prompt)
    else:
        batch_size = prompt_embeds.shape[0]

    # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
    # of the Imagen paper: https://arxiv.org/pdf/2205.11487.pdf . `guidance_scale = 1`
    # corresponds to doing no classifier free guidance.
    do_classifier_free_guidance = guidance_scale > 1.0

    # 3. Encode input prompt
    prompt_embeds = self._encode_prompt(
        prompt,
        num_waveforms_per_prompt,
        do_classifier_free_guidance,
        negative_prompt,
        prompt_embeds=prompt_embeds,
        negative_prompt_embeds=negative_prompt_embeds,
    )

    # 4. Prepare timesteps
    self.scheduler.set_timesteps(num_inference_steps)
    timesteps = self.scheduler.timesteps

    # 5. Prepare latent variables
    num_channels_latents = self.unet.config.in_channels
    latents = self.prepare_latents(
        batch_size * num_waveforms_per_prompt,
        num_channels_latents,
        height,
        prompt_embeds.dtype,
        generator,
        latents,
    )

    # 6. Prepare extra step kwargs
    extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)

    # 7. Denoising loop
    num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order
    with self.progress_bar(total=num_inference_steps) as progress_bar:
        for i, t in enumerate(timesteps):
            # expand the latents if we are doing classifier free guidance
            latent_model_input = ops.cat([latents] * 2) if do_classifier_free_guidance else latents
            latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)

            # predict the noise residual
            noise_pred = self.unet(
                latent_model_input,
                t,
                encoder_hidden_states=None,
                class_labels=prompt_embeds,
                cross_attention_kwargs=cross_attention_kwargs,
                return_dict=False,
            )[0]

            # perform guidance
            if do_classifier_free_guidance:
                noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
                noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)

            # compute the previous noisy sample x_t -> x_t-1
            latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs)[0]

            # call the callback, if provided
            if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
                progress_bar.update()
                if callback is not None and i % callback_steps == 0:
                    step_idx = i // getattr(self.scheduler, "order", 1)
                    callback(step_idx, t, latents)

    # 8. Post-processing
    if not output_type == "latent":
        latents = 1 / self.vae.config.scaling_factor * latents
        mel_spectrogram = self.vae.decode(latents)[0]
    else:
        return AudioPipelineOutput(audios=latents)

    audio = self.mel_spectrogram_to_waveform(mel_spectrogram)

    audio = audio[:, :original_waveform_length]

    # 9. Automatic scoring
    if num_waveforms_per_prompt > 1 and prompt is not None:
        audio = self.score_waveforms(
            text=prompt,
            audio=audio,
            num_waveforms_per_prompt=num_waveforms_per_prompt,
            dtype=prompt_embeds.dtype,
        )

    if output_type == "np":
        audio = audio.numpy()

    if not return_dict:
        return (audio,)

    return AudioPipelineOutput(audios=audio)