--- license: mit datasets: - mozilla-foundation/common_voice_17_0 - espnet/yodas - facebook/multilingual_librispeech language: - fr metrics: - wer pipeline_tag: automatic-speech-recognition tags: - asr - whisper model-index: - name: distil-large-v3-fr results: - task: name: Automatic Speech Recognition type: automatic-speech-recognition dataset: name: Common Voice 17.0, short-form (<= 30sec) type: mozilla-foundation/common_voice_17_0 config: fr split: test args: language: fr metrics: - name: WER type: wer value: 12.681 - task: name: Automatic Speech Recognition type: automatic-speech-recognition dataset: name: Multilingual LibriSpeech, short-form (<= 30sec) type: facebook/multilingual_librispeech config: french split: test args: language: fr metrics: - name: WER type: wer value: 5.865 - task: name: Automatic Speech Recognition type: automatic-speech-recognition dataset: name: VoxPopuli, short-form (<= 30sec) type: facebook/voxpopuli config: fr split: test args: language: fr metrics: - name: WER type: wer value: 10.851 - task: name: Automatic Speech Recognition type: automatic-speech-recognition dataset: name: Fleurs, short-form (<= 30sec) type: google/fleurs config: fr_fr split: test args: language: fr metrics: - name: WER type: wer value: 7.984 --- # Distil-Whisper: distil-large-v3-fr Distil-Whisper for English Automatic Speech Recognition (ASR) was proposed in the paper [Robust Knowledge Distillation via Large-Scale Pseudo Labelling](https://arxiv.org/abs/2311.00430). This is the knowledge distilled version of OpenAI's [Whisper large-v3](https://huggingface.co/openai/whisper-large-v3) for French ASR. The result is a distilled model that performs within **2% WER of whisper-large-v3** on out-of-distribution evaluation sets for both short-form and long form transcription. Moreover, it is **5.9x** faster than whisper-large-v3 and **1.3** times faster than the tiniest version of whisper while being uncomparably more accurate. | Model | Params (M) | Rel. Latency | Short-Form WER | Long-Form WER | | :--------------------- | :--------: | :----------: | :------------: | :-----------: | | whisper-tiny | 37.8 | 4.7 | 43.24 | 28.28 | | whisper-base | 72.6 | 3.7 | 30.48 | 19.23 | | whisper-small | 242 | 2.3 | 16.36 | 12.47 | | whisper-medium | 764 | 1.3 | 11.53 | 10.77 | | whisper-large-v3 | 1540 | 1.0 | 7.84 | 9.07 | | **distil-large-v3-fr** | **756** | **5.9** | **9.36** | **11.47** | *latencies benchmarked to generate 128 tokens on A100 40GB with a batch size of 1. More details about inference performances in [inference speed](#inference-speed) section. *WERs are averaged on the test sets. More details in [short-form](#short-form) and [long-form](#long-form) results sections. ## Transformers Usage distil-large-v3-fr is supported in the Hugging Face 🤗 Transformers library from version 4.41 onwards. To run the model, first install the latest version of Transformers. For this example, we'll also install 🤗 Datasets to load a toy audio dataset from the Hugging Face Hub: ```bash pip install --upgrade pip pip install --upgrade transformers accelerate datasets[audio] ``` ### Short-Form Transcription The model can be used with the [`pipeline`](https://huggingface.co/docs/transformers/main_classes/pipelines#transformers.AutomaticSpeechRecognitionPipeline) class to transcribe short-form audio files (< 30-seconds) as follows: ```python import torch from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor, pipeline from datasets import load_dataset device = "cuda:0" if torch.cuda.is_available() else "cpu" torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32 model_id = "eustlb/distil-large-v3-fr" model = AutoModelForSpeechSeq2Seq.from_pretrained( model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True, use_safetensors=True ) model.to(device) processor = AutoProcessor.from_pretrained(model_id) pipe = pipeline( "automatic-speech-recognition", model=model, tokenizer=processor.tokenizer, feature_extractor=processor.feature_extractor, max_new_tokens=128, torch_dtype=torch_dtype, device=device, ) dataset = load_dataset("google/fleurs", "fr_fr", split="train", streaming=True) sample = next(iter(dataset))["audio"] result = pipe(sample) print(result["text"]) ``` To transcribe a local audio file, simply pass the path to your audio file when you call the pipeline: ```diff - result = pipe(sample) + result = pipe("audio.mp3") ``` For segment-level timestamps, pass the argument `return_timestamps=True` and return the `"chunks"` output: ```python result = pipe(sample, return_timestamps=True) print(result["chunks"]) ```
For more control over the generation parameters, use the model + processor API directly: Ad-hoc generation arguments can be passed to `model.generate`, including `num_beams` for beam-search, `return_timestamps` for segment-level timestamps, and `prompt_ids` for prompting. See the [docstrings](https://huggingface.co/docs/transformers/en/model_doc/whisper#transformers.WhisperForConditionalGeneration.generate) for more details. ```python import torch from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor from datasets import Audio, load_dataset device = "cuda:0" if torch.cuda.is_available() else "cpu" torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32 model_id = "eustlb/distil-large-v3-fr" model = AutoModelForSpeechSeq2Seq.from_pretrained( model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True, use_safetensors=True ) model.to(device) processor = AutoProcessor.from_pretrained(model_id) dataset = load_dataset("google/fleurs", "fr_fr", split="train", streaming=True) dataset = dataset.cast_column("audio", Audio(processor.feature_extractor.sampling_rate)) sample = next(iter(dataset))["audio"] input_features = processor( sample["array"], sampling_rate=sample["sampling_rate"], return_tensors="pt" ).input_features input_features = input_features.to(device, dtype=torch_dtype) gen_kwargs = { "max_new_tokens": 128, "num_beams": 1, "return_timestamps": False, } pred_ids = model.generate(input_features, **gen_kwargs) pred_text = processor.batch_decode(pred_ids, skip_special_tokens=True, decode_with_timestamps=gen_kwargs["return_timestamps"]) print(pred_text) ```
### Sequential Long-Form distil-large-v3 is compatible with OpenAI's sequential long-form transcription algorithm. This algorithm uses a sliding window for buffered inference of long audio files (> 30-seconds), and returns more accurate transcriptions compared to the [chunked long-form algorithm](#chunked-long-form). The sequential long-form algorithm should be used in either of the following scenarios: 1. Transcription accuracy is the most important factor, and latency is less of a consideration 2. You are transcribing **batches** of long audio files, in which case the latency of sequential is comparable to chunked, while being up to 0.5% WER more accurate If you are transcribing single long audio files and latency is the most important factor, you should use the chunked algorithm described [below](#chunked-long-form). For a detailed explanation of the different algorithms, refer to Sections 5 of the [Distil-Whisper paper](https://arxiv.org/pdf/2311.00430.pdf). The [`pipeline`](https://huggingface.co/docs/transformers/main_classes/pipelines#transformers.AutomaticSpeechRecognitionPipeline) class can be used to transcribe long audio files with the sequential algorithm as follows: ```python import torch from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor, pipeline from datasets import load_dataset device = "cuda:0" if torch.cuda.is_available() else "cpu" torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32 model_id = "eustlb/distil-large-v3-fr" model = AutoModelForSpeechSeq2Seq.from_pretrained( model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True, use_safetensors=True ) model.to(device) processor = AutoProcessor.from_pretrained(model_id) pipe = pipeline( "automatic-speech-recognition", model=model, tokenizer=processor.tokenizer, feature_extractor=processor.feature_extractor, max_new_tokens=128, torch_dtype=torch_dtype, device=device, ) dataset = load_dataset("eustlb/french-long-form-test", split="test", streaming=True) sample = next(iter(dataset))["audio"] result = pipe(sample) print(result["text"]) ```
For more control over the generation parameters, use the model + processor API directly: ```python import torch from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor from datasets import Audio, load_dataset device = "cuda:0" if torch.cuda.is_available() else "cpu" torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32 model_id = "eustlb/distil-large-v3-fr" model = AutoModelForSpeechSeq2Seq.from_pretrained( model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True, use_safetensors=True ) model.to(device) processor = AutoProcessor.from_pretrained(model_id) dataset = load_dataset("eustlb/french-long-form-test", split="test", streaming=True) dataset = dataset.cast_column("audio", Audio(processor.feature_extractor.sampling_rate)) sample = next(iter(dataset))["audio"] inputs = processor( sample["array"], sampling_rate=sample["sampling_rate"], return_tensors="pt", truncation=False, padding="longest", return_attention_mask=True, ) inputs = inputs.to(device, dtype=torch_dtype) gen_kwargs = { "max_new_tokens": 448, "num_beams": 1, "condition_on_prev_tokens": False, "compression_ratio_threshold": 1.35, # zlib compression ratio threshold (in token space) "temperature": (0.0, 0.2, 0.4, 0.6, 0.8, 1.0), "logprob_threshold": -1.0, "no_speech_threshold": 0.6, "return_timestamps": True, } pred_ids = model.generate(**inputs, **gen_kwargs) pred_text = processor.batch_decode(pred_ids, skip_special_tokens=True, decode_with_timestamps=False) print(pred_text) ```
### Chunked Long-Form distil-large-v3-fr remains compatible with the Transformers chunked long-form algorithm. This algorithm should be used when a single large audio file is being transcribed and the fastest possible inference is required. In such circumstances, the chunked algorithm is up to 9x faster than OpenAI's sequential long-form implementation (see Table 7 of the [Distil-Whisper paper](https://arxiv.org/pdf/2311.00430.pdf)). To enable chunking, pass the `chunk_length_s` parameter to the `pipeline`. For distil-large-v3-fr, a chunk length of 25-seconds is optimal. To activate batching over long audio files, pass the argument `batch_size`: ```python import torch from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor, pipeline from datasets import load_dataset device = "cuda:0" if torch.cuda.is_available() else "cpu" torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32 model_id = "eustlb/distil-large-v3-fr" model = AutoModelForSpeechSeq2Seq.from_pretrained( model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True, use_safetensors=True ) model.to(device) processor = AutoProcessor.from_pretrained(model_id) pipe = pipeline( "automatic-speech-recognition", model=model, tokenizer=processor.tokenizer, feature_extractor=processor.feature_extractor, max_new_tokens=128, chunk_length_s=25, batch_size=16, torch_dtype=torch_dtype, device=device, ) dataset = load_dataset("eustlb/french-long-form-test", split="test", streaming=True) sample = next(iter(dataset))["audio"] result = pipe(sample) print(result["text"]) ``` ### Speculative Decoding distil-large-v3 is the first Distil-Whisper model that can be used as an assistant to Whisper large-v3 for [speculative decoding](https://huggingface.co/blog/whisper-speculative-decoding). Speculative decoding mathematically ensures that exactly the same outputs as Whisper are obtained, while being 2 times faster. This makes it the perfect drop-in replacement for existing Whisper pipelines, since the same outputs are guaranteed. In the following code-snippet, we load the assistant Distil-Whisper model standalone to the main Whisper pipeline. We then specify it as the "assistant model" for generation: ```python from transformers import pipeline, AutoModelForCausalLM, AutoModelForSpeechSeq2Seq, AutoProcessor import torch from datasets import load_dataset device = "cuda:0" if torch.cuda.is_available() else "cpu" torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32 assistant_model_id = "eustlb/distil-large-v3-fr" assistant_model = AutoModelForCausalLM.from_pretrained( assistant_model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True, use_safetensors=True ) assistant_model.to(device) model_id = "openai/whisper-large-v3" model = AutoModelForSpeechSeq2Seq.from_pretrained( model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True, use_safetensors=True ) model.to(device) processor = AutoProcessor.from_pretrained(model_id) pipe = pipeline( "automatic-speech-recognition", model=model, tokenizer=processor.tokenizer, feature_extractor=processor.feature_extractor, max_new_tokens=128, generate_kwargs={"assistant_model": assistant_model}, torch_dtype=torch_dtype, device=device, ) dataset = load_dataset("google/fleurs", "fr_fr", split="train", streaming=True) sample = next(iter(dataset))["audio"] result = pipe(sample) print(result["text"]) ``` For more details on speculative decoding, refer to the blog post [Speculative Decoding for 2x Faster Whisper Inference](https://huggingface.co/blog/whisper-speculative-decoding). ### Additional Speed & Memory Improvements You can apply additional speed and memory improvements to Distil-Whisper to further reduce the inference speed and VRAM requirements. These optimisations primarily target the attention kernel, swapping it from an eager implementation to a more efficient flash attention version. #### Flash Attention 2 We recommend using [Flash-Attention 2](https://huggingface.co/docs/transformers/main/en/perf_infer_gpu_one#flashattention-2) if your GPU allows for it. To do so, you first need to install [Flash Attention](https://github.com/Dao-AILab/flash-attention): ``` pip install flash-attn --no-build-isolation ``` Then pass `attn_implementation="flash_attention_2"` to `from_pretrained`: ```diff - model = AutoModelForSpeechSeq2Seq.from_pretrained(model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True, use_safetensors=True) + model = AutoModelForSpeechSeq2Seq.from_pretrained(model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True, use_safetensors=True, attn_implementation="flash_attention_2") ``` #### Torch Scale-Product-Attention (SDPA) If your GPU does not support Flash Attention, we recommend making use of PyTorch [scaled dot-product attention (SDPA)](https://pytorch.org/docs/stable/generated/torch.nn.functional.scaled_dot_product_attention.html). This attention implementation is activated **by default** for PyTorch versions 2.1.1 or greater. To check whether you have a compatible PyTorch version, run the following Python code snippet: ```python from transformers.utils import is_torch_sdpa_available print(is_torch_sdpa_available()) ``` If the above returns `True`, you have a valid version of PyTorch installed and SDPA is activated by default. If it returns `False`, you need to upgrade your PyTorch version according to the [official instructions](https://pytorch.org/get-started/locally/) Once a valid PyTorch version is installed, SDPA is activated by default. It can also be set explicitly by specifying `attn_implementation="sdpa"` as follows: ```diff - model = AutoModelForSpeechSeq2Seq.from_pretrained(model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True, use_safetensors=True) + model = AutoModelForSpeechSeq2Seq.from_pretrained(model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True, use_safetensors=True, attn_implementation="sdpa") ``` For more information about how to use the SDPA refer to the [Transformers SDPA documentation](https://huggingface.co/docs/transformers/en/perf_infer_gpu_one#pytorch-scaled-dot-product-attention). #### Torch compile Coming soon... #### 4-bit and 8-bit Inference Coming soon... ## Library Integrations ### Whisper.cpp distil-large-v3-fr can be run with the [Whisper.cpp](https://github.com/ggerganov/whisper.cpp) package with the original sequential long-form transcription algorithm. In a provisional benchmark on Mac M1, distil-large-v3 is over 5x faster than Whisper large-v3, while performing to within 0.8% WER over long-form audio. Steps for getting started: 1. Clone the Whisper.cpp repository: ``` git clone https://github.com/ggerganov/whisper.cpp.git cd whisper.cpp ```whis 2. Install the Hugging Face Hub Python package: ```bash pip install --upgrade huggingface_hub ``` And download the GGML weights for distil-large-v3 using the following Python snippet: ```python from huggingface_hub import hf_hub_download hf_hub_download(repo_id='eustlb/distil-large-v3-fr-ggml', filename='ggml-distil-large-v3-fr.bin', local_dir='./models') ``` Note that if you do not have a Python environment set-up, you can also download the weights directly with `wget`: ```bash wget https://huggingface.co/distil-whisper/distil-large-v3-ggml/resolve/main/ggml-distil-large-v3-fr.bin -P ./models ```` ### Transformers.js Distil-Whisper can be run completely in your web browser with [Transformers.js](http://github.com/xenova/transformers.js): 1. Install Transformers.js from [NPM](https://www.npmjs.com/package/@xenova/transformers): ```bash npm i @xenova/transformers ``` 2. Import the library and perform inference with the pipeline API. ```js import { pipeline } from '@xenova/transformers'; const transcriber = await pipeline('automatic-speech-recognition', 'eustlb/distil-large-v3-fr'); const url = 'https://huggingface.co/datasets/Xenova/transformers.js-docs/resolve/main/jfk.wav'; const output = await transcriber(url); // { text: " And so, my fellow Americans, ask not what your country can do for you. Ask what you can do for your country." } ``` Refer to the Transformers.js [docs](https://huggingface.co/docs/transformers.js/api/pipelines#module_pipelines.AutomaticSpeechRecognitionPipeline) for further information. ## Data distil-large-v3-fr is trained on 4,515 hours of audio data from three open-source, permissively licensed speech datasets on the Hugging Face Hub: | Dataset | Size / h | Speakers | Domain | Licence | | --------------------------------------------------------------------------------------------- | -------- | -------- | ------------------ | ----------------------------------------------------------- | | [Common Voice 17](https://huggingface.co/datasets/mozilla-foundation/common_voice_17_0) | 1,014 | unknown | Narrated Wikipedia | [CC0-1.0](https://choosealicense.com/licenses/cc0-1.0/) | | [MultiLingual LibriSpeech](https://huggingface.co/datasets/facebook/multilingual_librispeech) | 1,077 | 142 | Audiobook | [CC-BY-4.0](https://choosealicense.com/licenses/cc-by-4.0/) | | [YODAS fr000 split](https://huggingface.co/datasets/espnet/yodas) | 2,424 | unknown | YouTube | [CC-BY-3.0](https://creativecommons.org/licenses/by/3.0/) | | **Total** | 4,515 | 142+ | | | The audio data is then pseudo-labelled using the Whisper large-v3 model: we use Whisper to generate predictions for all the audio in our training set and use these as the target labels during training. Using pseudo-labels ensures that the transcriptions are consistently formatted across datasets and provides sequence-level distillation signal during training. ## WER Filter The Whisper pseudo-label predictions are subject to mis-transcriptions and hallucinations. To ensure we only train on accurate pseudo-labels, we employ a simple WER heuristic during training. First, we normalise the Whisper pseudo-labels and the ground truth labels provided by each dataset. We then compute the WER between these labels. If the WER exceeds a specified threshold, we discard the training example. Otherwise, we keep it for training. We chose for this training a WER threshold of 20%, resulting in an effective training set of 2110 hours (750 for [Common Voice 17](https://huggingface.co/datasets/mozilla-foundation/common_voice_17_0), 1040 for [MultiLingual LibriSpeech](https://huggingface.co/datasets/facebook/multilingual_librispeech) and 320 for [YODAS fr000 split](https://huggingface.co/datasets/espnet/yodas)). Section 9.2 of the [Distil-Whisper paper](https://arxiv.org/abs/2311.00430) demonstrates the effectiveness of this filter for improving downstream performance of the distilled model. We also partially attribute Distil-Whisper's robustness to hallucinations to this filter. ## Training The model was trained for 18,000 optimisation steps (or 14 epochs) with batch size 256. We saved the best model, based on the global wer score on validation splits, reached after 14,000 optimization steps (or 11 epochs). The two decoder layers were initialized from distil-large-v3 to leverage language transfer from English to French (more details [here](https://github.com/huggingface/distil-whisper/tree/main/training#22-language-transfer)). ## Results The distilled model performs to within 1% WER of Whisper large-v3 on out-of-distribution (Voxpopuli, Fleurs) short-form audio and within 2.5% WER on out-of-distribuion sequential long-form decoding. ### Short-Form | Model Name | RTF | Common Voice 17 | Multilingual Librispeech | Voxpopuli | Fleurs | | :----------------: | :-----: | :-------------: | :----------------------: | :-------: | :----: | | distil-large-v3-fr | 319.543 | 12.726 | 5.823 | 10.808 | 8.067 | | whisper-tiny | 280.576 | 56.757 | 37.512 | 32.505 | 46.173 | | whisper-base | 261.235 | 42.447 | 25.2 | 26.434 | 27.851 | | whisper-small | 249.676 | 22.469 | 14.097 | 14.61 | 14.283 | | whisper-medium | 170.9 | 15.432 | 9.602 | 11.92 | 9.155 | | whisper-large-v3 | 150.719 | 11.024 | 4.783 | 9.948 | 5.624 | *the above datasets correspond to test splits, RTF co ### Long-Form | Model Name | RTF | [long-form test set](https://huggingface.co/datasets/speech-recognition-community-v2/dev_data) | | :----------------: | :-----: | :--------------------------------------------------------------------------------------------: | | distil-large-v3-fr | 176.626 | 11.467 | | whisper-tiny | 125.367 | 28.277 | | whisper-base | 110.139 | 19.228 | | whisper-small | 83.417 | 12.467 | | whisper-medium | 56.677 | 10.772 | | whisper-large-v3 | 41.805 | 9.073 | ### Inference speed Reported latencies were benchmarked on a 40GB nvidia A100, generating 128 tokens with SDPA, bfloat16, 3 warmup steps, 5 measures, one beam. The benchmarking script can be found here. The time measured is the time do one forward pass of the encoder and 128 autoregressive forward passes of the decoder.

latencies

## Reproducing Distil-Whisper Training and evaluation code to reproduce Distil-Whisper is available under the Distil-Whisper repository: https://github.com/huggingface/distil-whisper/tree/main/training ## License distil-large-v3-fr inherits the [MIT license](https://github.com/huggingface/distil-whisper/blob/main/LICENSE) from OpenAI's Whisper model. ## Citation If you use this model, please consider citing the [Distil-Whisper paper](https://arxiv.org/abs/2311.00430): ``` @misc{gandhi2023distilwhisper, title={Distil-Whisper: Robust Knowledge Distillation via Large-Scale Pseudo Labelling}, author={Sanchit Gandhi and Patrick von Platen and Alexander M. Rush}, year={2023}, eprint={2311.00430}, archivePrefix={arXiv}, primaryClass={cs.CL} } ``` ## Acknowledgements * OpenAI for the Whisper [model](https://huggingface.co/openai/whisper-large-v3), in particular Jong Wook Kim for the [original codebase](https://github.com/openai/whisper) and training discussions * Hugging Face 🤗 [Transformers](https://github.com/huggingface/transformers) for the model integration * [Georgi Gerganov](https://huggingface.co/ggerganov) for the Whisper cpp integration * [Joshua Lochner](https://huggingface.co/xenova) for the Transformers.js integration * [Laurent Mazare](https://huggingface.co/lmz) for the Candle integration * [Vaibhav Srivastav](https://huggingface.co/reach-vb) for Distil-Whisper distribution * [Raghav Sonavane](https://huggingface.co/rsonavane/distil-whisper-large-v2-8-ls) for an early iteration of Distil-Whisper on the LibriSpeech dataset