Eustache Le Bihan commited on
Commit
8fe9ca7
1 Parent(s): 0c3f150
Files changed (3) hide show
  1. app.py +5 -5
  2. assets/example_1.wav +2 -2
  3. assets/example_2.wav +2 -2
app.py CHANGED
@@ -123,7 +123,7 @@ if __name__ == "__main__":
123
  "
124
  >
125
  <h1 style="font-weight: 900; margin-bottom: 7px; line-height: normal;">
126
- Whisper vs Distil-Whisper: Speed Comparison
127
  </h1>
128
  </div>
129
  </div>
@@ -131,14 +131,14 @@ if __name__ == "__main__":
131
  )
132
  gr.HTML(
133
  f"""
134
- <p><a href="https://huggingface.co/distil-whisper/distil-large-v3"> Distil-Whisper</a> is a distilled variant
135
  of the <a href="https://huggingface.co/openai/whisper-large-v3"> Whisper</a> model by OpenAI. Compared to Whisper,
136
- Distil-Whisper runs 6x faster with 50% fewer parameters, while performing to within 1% word error rate (WER) on
137
  out-of-distribution evaluation data.</p>
138
 
139
- <p>In this demo, we perform a speed comparison between Whisper and Distil-Whisper in order to test this claim.
140
  Both models use the <a href="https://huggingface.co/distil-whisper/distil-large-v3#chunked-long-form"> chunked long-form transcription algorithm</a>
141
- in 🤗 Transformers. To use Distil-Whisper yourself, check the code examples on the
142
  <a href="https://github.com/huggingface/distil-whisper#1-usage"> Distil-Whisper repository</a>. To ensure fair
143
  usage of the Space, we ask that audio file inputs are kept to < 30 mins.</p>
144
  """
 
123
  "
124
  >
125
  <h1 style="font-weight: 900; margin-bottom: 7px; line-height: normal;">
126
+ Whisper vs distil-large-v3-fr: Speed Comparison
127
  </h1>
128
  </div>
129
  </div>
 
131
  )
132
  gr.HTML(
133
  f"""
134
+ <p><a href="https://huggingface.co/eustlb/distil-large-v3-fr">distil-large-v3-fr</a> is a distilled variant
135
  of the <a href="https://huggingface.co/openai/whisper-large-v3"> Whisper</a> model by OpenAI. Compared to Whisper,
136
+ distil-large-v3 runs 6x faster with 50% fewer parameters, while performing to within 1% word error rate (WER) on
137
  out-of-distribution evaluation data.</p>
138
 
139
+ <p>In this demo, we perform a speed comparison between Whisper and distil-whisper-large-v3 in order to test this claim.
140
  Both models use the <a href="https://huggingface.co/distil-whisper/distil-large-v3#chunked-long-form"> chunked long-form transcription algorithm</a>
141
+ in 🤗 Transformers. To use distil-large-3-fr, check the code examples on the
142
  <a href="https://github.com/huggingface/distil-whisper#1-usage"> Distil-Whisper repository</a>. To ensure fair
143
  usage of the Space, we ask that audio file inputs are kept to < 30 mins.</p>
144
  """
assets/example_1.wav CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:27ec158234655831502db2018ef76912db66b93e8fcd1c0a769f2d191ea3550e
3
- size 737338
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1e938b9f81dea096ec7d3752e90afca8d370f7a461d3a08e1a559f4440ed055d
3
+ size 1963810
assets/example_2.wav CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:024ce48133b75402d422f2a340293375b31947b84728794173873451286fc922
3
- size 1286430
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:81fc0857f7fe11416ede431db713a02fdb787bbc049802fe74c791f3b44e5bf4
3
+ size 1920044