Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,124 @@
|
|
| 1 |
-
|
| 2 |
-
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
|
| 2 |
+
---
|
| 3 |
+
pretty_name: "Full Clone of Whisper by Argmax for Experimentation"
|
| 4 |
+
viewer: false
|
| 5 |
+
library_name: whisperkit
|
| 6 |
+
tags:
|
| 7 |
+
- whisper
|
| 8 |
+
- whisperkit
|
| 9 |
+
- coreml
|
| 10 |
+
- asr
|
| 11 |
+
- quantized
|
| 12 |
+
---
|
| 13 |
+
|
| 14 |
+
# Original Repo:
|
| 15 |
+
https://huggingface.co/argmaxinc/whisperkit-coreml
|
| 16 |
+
|
| 17 |
+
# WhisperKit Transcription Quality
|
| 18 |
+
|
| 19 |
+
|
| 20 |
+
|
| 21 |
+
## Dataset: `librispeech`
|
| 22 |
+
Short-form Audio (<30s/clip) - 5 hours of English audiobook clips
|
| 23 |
+
|
| 24 |
+
| | WER (↓) | QoI (↑) | File Size (MB) | Code Commit |
|
| 25 |
+
|:------------------------------------------------------------------------------------------------------------------------------|:--------------------------------------------------------------------------------------------------------------------------------------|----------:|-----------------:|:---------------------------------------------------------------|
|
| 26 |
+
| large-v2 (WhisperOpenAIAPI) | [2.35](https://hf.co/datasets/argmaxinc/whisperkit-evals/tree/main/WhisperOpenAIAPI/openai_whisper-large-v2/librispeech) | 100 | 3100 | N/A |
|
| 27 |
+
| [large-v2](https://hf.co/argmaxinc/whisperkit-coreml/tree/main/openai_whisper-large-v2) | [2.77](https://hf.co/datasets/argmaxinc/whisperkit-evals/tree/main/WhisperKit/openai_whisper-large-v2/librispeech) | 96.6 | 3100 | [Link](https://github.com/argmaxinc/WhisperKit/commit/2846fd9) |
|
| 28 |
+
| [large-v2_949MB](https://hf.co/argmaxinc/whisperkit-coreml/tree/main/openai_whisper-large-v2_949MB) | [2.4](https://hf.co/datasets/argmaxinc/whisperkit-evals/tree/main/WhisperKit/openai_whisper-large-v2_949MB/librispeech) | 94.6 | 949 | [Link](https://github.com/argmaxinc/WhisperKit/commit/eca4a2e) |
|
| 29 |
+
| [large-v2_turbo](https://hf.co/argmaxinc/whisperkit-coreml/tree/main/openai_whisper-large-v2_turbo) | [2.76](https://hf.co/datasets/argmaxinc/whisperkit-evals/tree/main/WhisperKit/openai_whisper-large-v2_turbo/librispeech) | 96.6 | 3100 | [Link](https://github.com/argmaxinc/WhisperKit/commit/2846fd9) |
|
| 30 |
+
| [large-v2_turbo_955MB](https://hf.co/argmaxinc/whisperkit-coreml/tree/main/openai_whisper-large-v2_turbo_955MB) | [2.41](https://hf.co/datasets/argmaxinc/whisperkit-evals/tree/main/WhisperKit/openai_whisper-large-v2_turbo_955MB/librispeech) | 94.6 | 955 | [Link](https://github.com/argmaxinc/WhisperKit/commit/cf75348) |
|
| 31 |
+
| [large-v3](https://hf.co/argmaxinc/whisperkit-coreml/tree/main/openai_whisper-large-v3) | [2.04](https://hf.co/datasets/argmaxinc/whisperkit-evals/tree/main/WhisperKit/openai_whisper-large-v3/librispeech) | 95.2 | 3100 | [Link](https://github.com/argmaxinc/WhisperKit/commit/2846fd9) |
|
| 32 |
+
| [large-v3_turbo](https://hf.co/argmaxinc/whisperkit-coreml/tree/main/openai_whisper-large-v3_turbo) | [2.03](https://hf.co/datasets/argmaxinc/whisperkit-evals/tree/main/WhisperKit/openai_whisper-large-v3_turbo/librispeech) | 95.4 | 3100 | [Link](https://github.com/argmaxinc/WhisperKit/commit/2846fd9) |
|
| 33 |
+
| [large-v3_turbo_954MB](https://hf.co/argmaxinc/whisperkit-coreml/tree/main/openai_whisper-large-v3_turbo_954MB) | [2.47](https://hf.co/datasets/argmaxinc/whisperkit-evals/tree/main/WhisperKit/openai_whisper-large-v3_turbo_954MB/librispeech) | 93.9 | 954 | [Link](https://github.com/argmaxinc/WhisperKit/commit/cf75348) |
|
| 34 |
+
| [distil-large-v3](https://hf.co/argmaxinc/whisperkit-coreml/tree/main/distil-whisper_distil-large-v3) | [2.47](https://hf.co/datasets/argmaxinc/whisperkit-evals/tree/main/WhisperKit/distil-whisper_distil-large-v3/librispeech) | 89.7 | 1510 | [Link](https://github.com/argmaxinc/WhisperKit/commit/cf75348) |
|
| 35 |
+
| [distil-large-v3_594MB](https://hf.co/argmaxinc/whisperkit-coreml/tree/main/distil-whisper_distil-large-v3_594MB) | [2.96](https://hf.co/datasets/argmaxinc/whisperkit-evals/tree/main/WhisperKit/distil-whisper_distil-large-v3_594MB/librispeech) | 85.4 | 594 | [Link](https://github.com/argmaxinc/WhisperKit/commit/508240f) |
|
| 36 |
+
| [distil-large-v3_turbo](https://hf.co/argmaxinc/whisperkit-coreml/tree/main/distil-whisper_distil-large-v3_turbo) | [2.47](https://hf.co/datasets/argmaxinc/whisperkit-evals/tree/main/WhisperKit/distil-whisper_distil-large-v3_turbo/librispeech) | 89.7 | 1510 | [Link](https://github.com/argmaxinc/WhisperKit/commit/508240f) |
|
| 37 |
+
| [distil-large-v3_turbo_600MB](https://hf.co/argmaxinc/whisperkit-coreml/tree/main/distil-whisper_distil-large-v3_turbo_600MB) | [2.78](https://hf.co/datasets/argmaxinc/whisperkit-evals/tree/main/WhisperKit/distil-whisper_distil-large-v3_turbo_600MB/librispeech) | 86.2 | 600 | [Link](https://github.com/argmaxinc/WhisperKit/commit/ae1cf96) |
|
| 38 |
+
| [small.en](https://hf.co/argmaxinc/whisperkit-coreml/tree/main/openai_whisper-small.en) | [3.12](https://hf.co/datasets/argmaxinc/whisperkit-evals/tree/main/WhisperKit/openai_whisper-small.en/librispeech) | 85.8 | 483 | [Link](https://github.com/argmaxinc/WhisperKit/commit/228630c) |
|
| 39 |
+
| [small](https://hf.co/argmaxinc/whisperkit-coreml/tree/main/openai_whisper-small) | [3.45](https://hf.co/datasets/argmaxinc/whisperkit-evals/tree/main/WhisperKit/openai_whisper-small/librispeech) | 83 | 483 | [Link](https://github.com/argmaxinc/WhisperKit/commit/228630c) |
|
| 40 |
+
| [base.en](https://hf.co/argmaxinc/whisperkit-coreml/tree/main/openai_whisper-base.en) | [3.98](https://hf.co/datasets/argmaxinc/whisperkit-evals/tree/main/WhisperKit/openai_whisper-base.en/librispeech) | 75.3 | 145 | [Link](https://github.com/argmaxinc/WhisperKit/commit/228630c) |
|
| 41 |
+
| [base](https://hf.co/argmaxinc/whisperkit-coreml/tree/main/openai_whisper-base) | [4.97](https://hf.co/datasets/argmaxinc/whisperkit-evals/tree/main/WhisperKit/openai_whisper-base/librispeech) | 67.2 | 145 | [Link](https://github.com/argmaxinc/WhisperKit/commit/228630c) |
|
| 42 |
+
| [tiny.en](https://hf.co/argmaxinc/whisperkit-coreml/tree/main/openai_whisper-tiny.en) | [5.61](https://hf.co/datasets/argmaxinc/whisperkit-evals/tree/main/WhisperKit/openai_whisper-tiny.en/librispeech) | 63.9 | 66 | [Link](https://github.com/argmaxinc/WhisperKit/commit/228630c) |
|
| 43 |
+
| [tiny](https://hf.co/argmaxinc/whisperkit-coreml/tree/main/openai_whisper-tiny) | [7.47](https://hf.co/datasets/argmaxinc/whisperkit-evals/tree/main/WhisperKit/openai_whisper-tiny/librispeech) | 52.5 | 66 | [Link](https://github.com/argmaxinc/WhisperKit/commit/228630c) |
|
| 44 |
+
|
| 45 |
+
## Dataset: `earnings22`
|
| 46 |
+
Long-Form Audio (>1hr/clip) - 120 hours of earnings call recordings in English with various accents
|
| 47 |
+
|
| 48 |
+
| | WER (↓) | QoI (↑) | File Size (MB) | Code Commit |
|
| 49 |
+
|:------------------------------------------------------------------------------------------------------|:--------------------------------------------------------------------------------------------------------------------------|----------:|-----------------:|:---------------------------------------------------------------|
|
| 50 |
+
| large-v2 (WhisperOpenAIAPI) | [16.27](https://hf.co/datasets/argmaxinc/whisperkit-evals/tree/main/WhisperOpenAIAPI/openai_whisper-large-v2/earnings22) | 100 | 3100 | N/A |
|
| 51 |
+
| [large-v3](https://hf.co/argmaxinc/whisperkit-coreml/tree/main/openai_whisper-large-v3) | [15.17](https://hf.co/datasets/argmaxinc/whisperkit-evals/tree/main/WhisperKit/openai_whisper-large-v3/earnings22) | 58.5 | 3100 | [Link](https://github.com/argmaxinc/WhisperKit/commit/2846fd9) |
|
| 52 |
+
| [distil-large-v3](https://hf.co/argmaxinc/whisperkit-coreml/tree/main/distil-whisper_distil-large-v3) | [15.28](https://hf.co/datasets/argmaxinc/whisperkit-evals/tree/main/WhisperKit/distil-whisper_distil-large-v3/earnings22) | 46.3 | 1510 | [Link](https://github.com/argmaxinc/WhisperKit/commit/508240f) |
|
| 53 |
+
| [base.en](https://hf.co/argmaxinc/whisperkit-coreml/tree/main/openai_whisper-base.en) | [23.49](https://hf.co/datasets/argmaxinc/whisperkit-evals/tree/main/WhisperKit/openai_whisper-base.en/earnings22) | 6.5 | 145 | [Link](https://github.com/argmaxinc/WhisperKit/commit/dda6571) |
|
| 54 |
+
| [tiny.en](https://hf.co/argmaxinc/whisperkit-coreml/tree/main/openai_whisper-tiny.en) | [28.64](https://hf.co/datasets/argmaxinc/whisperkit-evals/tree/main/WhisperKit/openai_whisper-tiny.en/earnings22) | 5.7 | 66 | [Link](https://github.com/argmaxinc/WhisperKit/commit/dda6571) |
|
| 55 |
+
|
| 56 |
+
|
| 57 |
+
### Explanation
|
| 58 |
+
|
| 59 |
+
We believe that rigorously measuring the quality of inference is necessary for developers and
|
| 60 |
+
enterprises to make informed decisions when opting to use optimized or compressed variants of
|
| 61 |
+
any machine learning model in production. To contextualize `WhisperKit`, we take the following Whisper
|
| 62 |
+
implementations and benchmark them using a consistent evaluation harness:
|
| 63 |
+
|
| 64 |
+
Server-side:
|
| 65 |
+
- `WhisperOpenAIAPI`: [OpenAI's Whisper API](https://platform.openai.com/docs/guides/speech-to-text)
|
| 66 |
+
|
| 67 |
+
($0.36 per hour of audio as of 02/29/24, 25MB file size limit per request)
|
| 68 |
+
|
| 69 |
+
On-device:
|
| 70 |
+
- `WhisperKit`: Argmax's implementation [[Eval Harness]](https://github.com/argmaxinc/whisperkittools/blob/main/whisperkit/pipelines.py#L100) [[Repo]](https://github.com/argmaxinc/WhisperKit)
|
| 71 |
+
- `whisper.cpp`: A C++ implementation form ggerganov [[Eval Harness]](https://github.com/argmaxinc/whisperkittools/blob/main/whisperkit/pipelines.py#L212) [[Repo]](https://github.com/ggerganov/whisper.cpp)
|
| 72 |
+
- `WhisperMLX`: A Python implementation from Apple MLX [[Eval Harness]](https://github.com/argmaxinc/whisperkittools/blob/main/whisperkit/pipelines.py#L338) [[Repo]](https://github.com/ml-explore/mlx-examples/blob/main/whisper/whisper/transcribe.py)
|
| 73 |
+
|
| 74 |
+
(All on-device implementations are available for free under MIT license as of 03/19/2024)
|
| 75 |
+
|
| 76 |
+
`WhisperOpenAIAPI` sets the reference and we assume that it is using the equivalent of [openai/whisper-large-v2](https://huggingface.co/openai/whisper-large-v2)
|
| 77 |
+
in float16 precision along with additional undisclosed optimizations from OpenAI. In all measurements, we care primarily about per-example no-regressions (quantified as `qoi` below)
|
| 78 |
+
which is a stricter metric compared to dataset average [Word Error RATE (WER)](https://en.wikipedia.org/wiki/Word_error_rate). A 100% `qoi` preserves perfect backwards-compatibility on the test distribution and avoids "perceived regressions", the phenomenon
|
| 79 |
+
where per-example known behavior changes after a code/model update and causes divergence in downstream code or breaks the user experience itself (even if dataset averages might stay flat
|
| 80 |
+
across updates). Pseudocode for `qoi`:
|
| 81 |
+
|
| 82 |
+
```python
|
| 83 |
+
qoi = []
|
| 84 |
+
for example in dataset:
|
| 85 |
+
no_regression = wer(optimized_model(example)) <= wer(reference_model(example))
|
| 86 |
+
qoi.append(no_regression)
|
| 87 |
+
qoi = (sum(qoi) / len(qoi)) * 100.
|
| 88 |
+
```
|
| 89 |
+
|
| 90 |
+
Note that the ordering of models with respect to `WER` does not necessarily match the ordering with respect to `QoI`. This is because the reference model gets assigned
|
| 91 |
+
a QoI of 100% by definition. Any per-example regression by other implementations get penalized while per-example improvements are not rewarded. `QoI` (higher is better) matters
|
| 92 |
+
where the production behavior is established by the reference results and the goal is to not regress when switching to an optimized or compressed model. On the other hand,
|
| 93 |
+
`WER` (lower is better) matters when there is no established production behavior and one is picking the best quality versus model size trade off point.
|
| 94 |
+
|
| 95 |
+
We anticipate developers that use Whisper (or similar models) in production to have their own Quality Assurance test sets and [whisperkittools](https://github.com/argmaxinc/whisperkittools) offers
|
| 96 |
+
the tooling necessary to run the same measurements on such custom test sets, please see the [Model Evaluation on Custom Dataset]((https://github.com/argmaxinc/whisperkittools)) for details.
|
| 97 |
+
|
| 98 |
+
### Why are there so many Whisper versions?
|
| 99 |
+
WhisperKit is an SDK for building speech-to-text features in apps across a wide range of Apple devices. We are working towards abstracting away the model versioning from the developer so WhisperKit
|
| 100 |
+
"just works" by deploying the highest-quality model version that a particular device can execute. In the interim, we leave the choice to the developer by providing quality and size trade-offs.
|
| 101 |
+
|
| 102 |
+
|
| 103 |
+
### Datasets
|
| 104 |
+
- [librispeech](https://huggingface.co/datasets/argmaxinc/librispeech): ~5 hours of short English audio clips, tests short-form transcription quality
|
| 105 |
+
- [earnings22](https://huggingface.co/datasets/argmaxinc/earnings22): ~120 hours of English audio clips from earnings calls with various accents, tests long-form transcription quality
|
| 106 |
+
|
| 107 |
+
### Reproducing Results
|
| 108 |
+
Benchmark results on this page were automatically generated by [whisperkittools](https://github.com/argmaxinc/whisperkittools) using our cluster of Apple Silicon Macs as self-hosted runners on
|
| 109 |
+
Github Actions. We periodically recompute these benchmarks as part of our CI pipeline. Due to [security concerns](https://docs.github.com/en/actions/security-guides/security-hardening-for-github-actions#hardening-for-self-hosted-runners),
|
| 110 |
+
we are unable to open up the cluster to the public. However, any Apple Silicon Mac (even with 8GB RAM) can be used to
|
| 111 |
+
run identical [evaluation jobs](#evaluation) locally. For reference, our M2 Ultra devices complete a `librispeech` + `openai/whisper-large-v3`
|
| 112 |
+
evaluation in under 1 hour regardless of the Whisper implementation. Oldest Apple Silicon Macs should take less than 1 day to complete the same evaluation.
|
| 113 |
+
|
| 114 |
+
|
| 115 |
+
|
| 116 |
+
### Glossary
|
| 117 |
+
|
| 118 |
+
- `_turbo`: Indicates the presence of additional optimizations (not compression) to unlock streaming transcription
|
| 119 |
+
as described in our [Blog Post](https://www.takeargmax.com/blog/whisperkit).
|
| 120 |
+
|
| 121 |
+
- `_*MB`: Indicates the presence of model compression. Instead of cluttering the filename with details like
|
| 122 |
+
`_AudioEncoder-5.8bits_TextDecoder-6.1bits_QLoRA-rank=16`, we choose to summarize the compression spec as the
|
| 123 |
+
resulting total file size since this is what matters to developers in production.
|
| 124 |
+
|