readme
Browse files
README.md
CHANGED
@@ -1,201 +1,183 @@
|
|
1 |
---
|
2 |
-
|
3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
4 |
---
|
5 |
|
6 |
-
#
|
7 |
-
|
8 |
-
|
9 |
-
|
10 |
-
|
11 |
-
|
12 |
-
|
13 |
-
|
14 |
-
|
15 |
-
|
16 |
-
|
17 |
-
|
18 |
-
|
19 |
-
|
20 |
-
|
21 |
-
|
22 |
-
|
23 |
-
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
|
28 |
-
|
29 |
-
|
30 |
-
|
31 |
-
|
32 |
-
|
33 |
-
|
34 |
-
|
35 |
-
|
36 |
-
|
37 |
-
|
38 |
-
|
39 |
-
|
40 |
-
|
41 |
-
|
42 |
-
|
43 |
-
|
44 |
-
|
45 |
-
|
46 |
-
|
47 |
-
|
48 |
-
|
49 |
-
|
50 |
-
|
51 |
-
|
52 |
-
|
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 |
-
[More Information Needed]
|
130 |
-
|
131 |
-
#### Summary
|
132 |
-
|
133 |
-
|
134 |
-
|
135 |
-
## Model Examination [optional]
|
136 |
-
|
137 |
-
<!-- Relevant interpretability work for the model goes here -->
|
138 |
-
|
139 |
-
[More Information Needed]
|
140 |
-
|
141 |
-
## Environmental Impact
|
142 |
-
|
143 |
-
<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
|
144 |
-
|
145 |
-
Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
|
146 |
-
|
147 |
-
- **Hardware Type:** [More Information Needed]
|
148 |
-
- **Hours used:** [More Information Needed]
|
149 |
-
- **Cloud Provider:** [More Information Needed]
|
150 |
-
- **Compute Region:** [More Information Needed]
|
151 |
-
- **Carbon Emitted:** [More Information Needed]
|
152 |
-
|
153 |
-
## Technical Specifications [optional]
|
154 |
-
|
155 |
-
### Model Architecture and Objective
|
156 |
-
|
157 |
-
[More Information Needed]
|
158 |
-
|
159 |
-
### Compute Infrastructure
|
160 |
-
|
161 |
-
[More Information Needed]
|
162 |
-
|
163 |
-
#### Hardware
|
164 |
-
|
165 |
-
[More Information Needed]
|
166 |
-
|
167 |
-
#### Software
|
168 |
-
|
169 |
-
[More Information Needed]
|
170 |
-
|
171 |
-
## Citation [optional]
|
172 |
-
|
173 |
-
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
|
174 |
-
|
175 |
-
**BibTeX:**
|
176 |
-
|
177 |
-
[More Information Needed]
|
178 |
-
|
179 |
-
**APA:**
|
180 |
-
|
181 |
-
[More Information Needed]
|
182 |
-
|
183 |
-
## Glossary [optional]
|
184 |
-
|
185 |
-
<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
|
186 |
-
|
187 |
-
[More Information Needed]
|
188 |
-
|
189 |
-
## More Information [optional]
|
190 |
-
|
191 |
-
[More Information Needed]
|
192 |
-
|
193 |
-
## Model Card Authors [optional]
|
194 |
-
|
195 |
-
[More Information Needed]
|
196 |
-
|
197 |
-
## Model Card Contact
|
198 |
-
|
199 |
-
[More Information Needed]
|
200 |
-
|
201 |
-
|
|
|
1 |
---
|
2 |
+
license: apache-2.0
|
3 |
+
language: ja
|
4 |
+
tags:
|
5 |
+
- audio
|
6 |
+
- automatic-speech-recognition
|
7 |
+
- hf-asr-leaderboard
|
8 |
+
widget:
|
9 |
+
- example_title: CommonVoice 8.0 (Test Split)
|
10 |
+
src: >-
|
11 |
+
https://huggingface.co/datasets/japanese-asr/ja_asr.common_voice_8_0/resolve/main/sample.flac
|
12 |
+
- example_title: JSUT Basic 5000
|
13 |
+
src: >-
|
14 |
+
https://huggingface.co/datasets/japanese-asr/ja_asr.jsut_basic5000/resolve/main/sample.flac
|
15 |
+
- example_title: ReazonSpeech (Test Split)
|
16 |
+
src: >-
|
17 |
+
https://huggingface.co/datasets/japanese-asr/ja_asr.reazonspeech_test/resolve/main/sample.flac
|
18 |
+
pipeline_tag: automatic-speech-recognition
|
19 |
+
metrics:
|
20 |
+
- wer
|
21 |
+
model-index:
|
22 |
+
- name: kotoba-tech/kotoba-whisper-v1.0
|
23 |
+
results:
|
24 |
+
- task:
|
25 |
+
type: automatic-speech-recognition
|
26 |
+
dataset:
|
27 |
+
name: CommonVoice_8.0 (Japanese)
|
28 |
+
type: japanese-asr/ja_asr.common_voice_8_0
|
29 |
+
metrics:
|
30 |
+
- name: WER
|
31 |
+
type: WER
|
32 |
+
value: TBA
|
33 |
+
- name: CER
|
34 |
+
type: CER
|
35 |
+
value: TBA
|
36 |
+
- task:
|
37 |
+
type: automatic-speech-recognition
|
38 |
+
dataset:
|
39 |
+
name: ReazonSpeech (Test)
|
40 |
+
type: japanese-asr/ja_asr.reazonspeech_test
|
41 |
+
metrics:
|
42 |
+
- name: WER
|
43 |
+
type: WER
|
44 |
+
value: TBA
|
45 |
+
- name: CER
|
46 |
+
type: CER
|
47 |
+
value: TBA
|
48 |
+
- task:
|
49 |
+
type: automatic-speech-recognition
|
50 |
+
dataset:
|
51 |
+
name: JSUT Basic5000
|
52 |
+
type: japanese-asr/ja_asr.jsut_basic5000
|
53 |
+
metrics:
|
54 |
+
- name: WER
|
55 |
+
type: WER
|
56 |
+
value: TBA
|
57 |
+
- name: CER
|
58 |
+
type: CER
|
59 |
+
value: TBA
|
60 |
---
|
61 |
|
62 |
+
# Kotoba-Whisper-v1.1
|
63 |
+
_Kotoba-Whisper-v1.1_ is a Japanese ASR model based on [kotoba-tech/kotoba-whisper-v1.0](https://huggingface.co/kotoba-tech/kotoba-whisper-v1.0), with
|
64 |
+
additional postprocessing stacks integrated as [`pipeline`](https://huggingface.co/docs/transformers/en/main_classes/pipelines). The new features includes
|
65 |
+
(i) improved timestamp achieved by [stable-ts](https://github.com/jianfch/stable-ts) and (ii) adding punctuation with [punctuators](https://github.com/1-800-BAD-CODE/punctuators/tree/main).
|
66 |
+
These libraries are merged into Kotoba-Whisper-v1.1 via pipeline and will be applied seamlessly to the predicted transcription from [kotoba-tech/kotoba-whisper-v1.0](https://huggingface.co/kotoba-tech/kotoba-whisper-v1.0).
|
67 |
+
The pipeline has been developed through the collaboration between [Asahi Ushio](https://asahiushio.com) and [Kotoba Technologies](https://twitter.com/kotoba_tech)
|
68 |
+
|
69 |
+
## Transformers Usage
|
70 |
+
Kotoba-Whisper-v1.1 is supported in the Hugging Face π€ Transformers library from version 4.39 onwards. To run the model, first
|
71 |
+
install the latest version of Transformers.
|
72 |
+
|
73 |
+
```bash
|
74 |
+
pip install --upgrade pip
|
75 |
+
pip install --upgrade transformers accelerate
|
76 |
+
```
|
77 |
+
|
78 |
+
### Transcription
|
79 |
+
The model can be used with the [`pipeline`](https://huggingface.co/docs/transformers/main_classes/pipelines#transformers.AutomaticSpeechRecognitionPipeline)
|
80 |
+
class to transcribe audio files as follows:
|
81 |
+
|
82 |
+
```python
|
83 |
+
import torch
|
84 |
+
from transformers import pipeline
|
85 |
+
from datasets import load_dataset
|
86 |
+
|
87 |
+
# config
|
88 |
+
model_id = "kotoba-tech/kotoba-whisper-v1.1"
|
89 |
+
torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32
|
90 |
+
device = "cuda:0" if torch.cuda.is_available() else "cpu"
|
91 |
+
model_kwargs = {"attn_implementation": "sdpa"} if torch.cuda.is_available() else {}
|
92 |
+
generate_kwargs = {"language": "japanese", "task": "transcribe"}
|
93 |
+
|
94 |
+
# load model
|
95 |
+
pipe = pipeline(
|
96 |
+
model=model_id,
|
97 |
+
torch_dtype=torch_dtype,
|
98 |
+
device=device,
|
99 |
+
model_kwargs=model_kwargs,
|
100 |
+
chunk_length_s=15,
|
101 |
+
batch_size=16
|
102 |
+
)
|
103 |
+
|
104 |
+
# load sample audio
|
105 |
+
dataset = load_dataset("japanese-asr/ja_asr.reazonspeech_test", split="test")
|
106 |
+
sample = dataset[0]["audio"]
|
107 |
+
|
108 |
+
# run inference
|
109 |
+
result = pipe(sample, return_timestamps=True, generate_kwargs=generate_kwargs)
|
110 |
+
print(result)
|
111 |
+
```
|
112 |
+
|
113 |
+
- To transcribe a local audio file, simply pass the path to your audio file when you call the pipeline:
|
114 |
+
```diff
|
115 |
+
- result = pipe(sample, return_timestamps=True, generate_kwargs=generate_kwargs)
|
116 |
+
+ result = pipe("audio.mp3", return_timestamps=True, generate_kwargs=generate_kwargs)
|
117 |
+
```
|
118 |
+
|
119 |
+
### Transcription with Prompt
|
120 |
+
Kotoba-whisper can generate transcription with prompting as below:
|
121 |
+
|
122 |
+
```python
|
123 |
+
import re
|
124 |
+
import torch
|
125 |
+
from transformers import pipeline
|
126 |
+
from datasets import load_dataset
|
127 |
+
|
128 |
+
# config
|
129 |
+
model_id = "kotoba-tech/kotoba-whisper-v1.1"
|
130 |
+
torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32
|
131 |
+
device = "cuda:0" if torch.cuda.is_available() else "cpu"
|
132 |
+
model_kwargs = {"attn_implementation": "sdpa"} if torch.cuda.is_available() else {}
|
133 |
+
generate_kwargs = {"language": "japanese", "task": "transcribe"}
|
134 |
+
|
135 |
+
# load model
|
136 |
+
pipe = pipeline(
|
137 |
+
model=model_id,
|
138 |
+
torch_dtype=torch_dtype,
|
139 |
+
device=device,
|
140 |
+
model_kwargs=model_kwargs,
|
141 |
+
chunk_length_s=15,
|
142 |
+
batch_size=16
|
143 |
+
)
|
144 |
+
|
145 |
+
# load sample audio
|
146 |
+
dataset = load_dataset("japanese-asr/ja_asr.reazonspeech_test", split="test")
|
147 |
+
|
148 |
+
# --- Without prompt ---
|
149 |
+
text = pipe(dataset[10]["audio"], generate_kwargs=generate_kwargs)['text']
|
150 |
+
print(text)
|
151 |
+
# 81ζ³γεεΌ·γθ΅°γγ«ε€γγ£γ¦γγΎγγ
|
152 |
+
|
153 |
+
# --- With prompt ---: Let's change `81` to `91`.
|
154 |
+
prompt = "91ζ³"
|
155 |
+
generate_kwargs['prompt_ids'] = pipe.tokenizer.get_prompt_ids(prompt, return_tensors="pt").to(device)
|
156 |
+
text = pipe(dataset[10]["audio"], generate_kwargs=generate_kwargs)['text']
|
157 |
+
# currently the pipeline for ASR appends the prompt at the beginning of the transcription, so remove it
|
158 |
+
text = re.sub(rf"\A\s*{prompt}\s*", "", text)
|
159 |
+
print(text)
|
160 |
+
# γγ£γΆγ£γγ§γγΉγ«γ¬γγγ91ζ³γεεΌ·γθ΅°γγ«ε€γγ£γ¦γγΎγγ
|
161 |
+
```
|
162 |
+
|
163 |
+
### Flash Attention 2
|
164 |
+
We recommend using [Flash-Attention 2](https://huggingface.co/docs/transformers/main/en/perf_infer_gpu_one#flashattention-2)
|
165 |
+
if your GPU allows for it. To do so, you first need to install [Flash Attention](https://github.com/Dao-AILab/flash-attention):
|
166 |
+
|
167 |
+
```
|
168 |
+
pip install flash-attn --no-build-isolation
|
169 |
+
```
|
170 |
+
|
171 |
+
Then pass `attn_implementation="flash_attention_2"` to `from_pretrained`:
|
172 |
+
|
173 |
+
```diff
|
174 |
+
- model_kwargs = {"attn_implementation": "sdpa"} if torch.cuda.is_available() else {}
|
175 |
+
+ model_kwargs = {"attn_implementation": "flash_attention_2"} if torch.cuda.is_available() else {}
|
176 |
+
```
|
177 |
+
|
178 |
+
|
179 |
+
## Acknowledgements
|
180 |
+
* [OpenAI](https://openai.com/) for the Whisper [model](https://huggingface.co/openai/whisper-large-v3).
|
181 |
+
* Hugging Face π€ [Transformers](https://github.com/huggingface/transformers) for the model integration.
|
182 |
+
* Hugging Face π€ for the [Distil-Whisper codebase](https://github.com/huggingface/distil-whisper).
|
183 |
+
* [Reazon Human Interaction Lab](https://research.reazon.jp/) for the [ReazonSpeech dataset](https://huggingface.co/datasets/reazon-research/reazonspeech).
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|