Hugging Face's logo

Amirhossein75
/
speech-intensity-whisper

Automatic Speech Recognition
Transformers
Safetensors
PyTorch
English
whisper
asr
multitask
intensity-regression
loudness
lufs
dbfs
pyloudnorm
jiwer
gradio
sagemaker
custom_code
Model card Files
xet
 Community
Amirhossein75 commited on
Commit
3142eae
Β·
1 Parent(s): 88915cc

update readme with metrics, add new model weights

Browse files
Files changed (4) hide show
  1. README.md +83 -0
  2. model.safetensors +1 -1
  3. test_metrics.json +6 -6
  4. training_args.bin +1 -1
README.md CHANGED
@@ -141,6 +141,89 @@ total_loss = asr_ce_loss + Ξ» * mse(intensity)
141
  - **ASR:** WER (via `jiwer`)
142
  - **Intensity:** RMSE in **dBFS** or **LUFS**
143
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
144
  ### Results
145
  - Example logs and a sample checkpoint are referenced in the repository (`training-test-logs/` and the README link). Reproduce numbers with the provided scripts for your environment.
146
 
 
141
  - **ASR:** WER (via `jiwer`)
142
  - **Intensity:** RMSE in **dBFS** or **LUFS**
143
 
144
+
145
+ ## πŸ“Š Results & Metrics
146
+
147
+ ### πŸ”Ž Highlights
148
+ - **Test WER (↓):** **4.6976**
149
+ - **Test Intensity RMSE (↓):** **0.7334**
150
+ - **Validation WER (↓):** 4.6973 β€’ **Validation Intensity RMSE (↓):** 1.4492
151
+
152
+ > Lower is better (↓). WER computed with `jiwer`. Intensity RMSE is the regression error on the loudness target (RMS dBFS by default, or LUFS if `--intensity_method lufs` is used).
153
+
154
+ ---
155
+
156
+ ### βœ… Full Metrics
157
+
158
+ #### Validation (Dev)
159
+ | Metric | Value |
160
+ |---|---|
161
+ | **Loss** | **2.2288** |
162
+ | **WER (↓)** | **4.6973** |
163
+ | **Intensity RMSE (↓)** | **1.4492** |
164
+ | **Runtime (s)** | **1,156.757** _(β‰ˆ 19mβ€―17s)_ |
165
+ | **Samples / s** | **2.337** |
166
+ | **Steps / s** | **0.292** |
167
+ | **Epoch** | **1** |
168
+
169
+ #### Test
170
+ | Metric | Value |
171
+ |---|---|
172
+ | **Loss** | **0.6631** |
173
+ | **WER (↓)** | **4.6976** |
174
+ | **Intensity RMSE (↓)** | **0.7334** |
175
+ | **Runtime (s)** | **1,129.272** _(β‰ˆ 18mβ€―49s)_ |
176
+ | **Samples / s** | **2.320** |
177
+ | **Steps / s** | **0.290** |
178
+ | **Epoch** | **1** |
179
+
180
+ #### Training Summary
181
+ | Metric | Value |
182
+ |---|---|
183
+ | **Train Loss** | **72.5232** |
184
+ | **Runtime (s)** | **6,115.966** _(β‰ˆ 1hβ€―41mβ€―56s)_ |
185
+ | **Samples / s** | **4.666** |
186
+ | **Steps / s** | **0.292** |
187
+ | **Epochs** | **1** |
188
+
189
+ ---
190
+
191
+ <details>
192
+ <summary><strong>Raw metrics (for reproducibility)</strong></summary>
193
+
194
+ ```json
195
+ {
196
+ "validation": {
197
+ "eval_loss": 2.228771209716797,
198
+ "eval_wer": 4.69732730414323,
199
+ "eval_intensity_rmse": 1.4492216110229492,
200
+ "eval_runtime": 1156.7567,
201
+ "eval_samples_per_second": 2.337,
202
+ "eval_steps_per_second": 0.292,
203
+ "epoch": 1.0
204
+ },
205
+ "training": {
206
+ "train_loss": 72.52319664163974,
207
+ "train_runtime": 6115.9656,
208
+ "train_samples_per_second": 4.666,
209
+ "train_steps_per_second": 0.292,
210
+ "epoch": 1.0
211
+ },
212
+ "test": {
213
+ "test_loss": 0.6630592346191406,
214
+ "test_wer": 4.69758064516129,
215
+ "test_intensity_rmse": 0.7333692312240601,
216
+ "test_runtime": 1129.2724,
217
+ "test_samples_per_second": 2.32,
218
+ "test_steps_per_second": 0.29,
219
+ "epoch": 1.0
220
+ }
221
+ }
222
+ ```
223
+
224
+ </details>
225
+
226
+
227
  ### Results
228
  - Example logs and a sample checkpoint are referenced in the repository (`training-test-logs/` and the README link). Reproduce numbers with the provided scripts for your environment.
229
 
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:d9a70a81969cc280b6d97501183ebe9129d43a0dc579b8063bd535bd3f067f62
3
  size 969367188
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8383a9108b6f4cb1e7ed8242d2296d9c946720622697f61fe5aa7290afffcdc2
3
  size 969367188
test_metrics.json CHANGED
@@ -1,9 +1,9 @@
1
  {
2
- "test_loss": 0.6354733109474182,
3
- "test_wer": 4.029785453438832,
4
- "test_runtime": 1117.4878,
5
- "test_samples_per_second": 2.345,
6
- "test_steps_per_second": 0.294,
7
  "epoch": 1.0,
8
- "test_intensity_rmse": 0.7143185138702393
9
  }
 
1
  {
2
+ "test_loss": 0.6630592346191406,
3
+ "test_wer": 4.69758064516129,
4
+ "test_runtime": 1129.2724,
5
+ "test_samples_per_second": 2.32,
6
+ "test_steps_per_second": 0.29,
7
  "epoch": 1.0,
8
+ "test_intensity_rmse": 0.7333692312240601
9
  }
training_args.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:8d95f6288fad556ad155510df559606ad2964fdf6440a7121a2f7cbf9071c2d5
3
  size 5905
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ed453f968ff1de2f17e35993de55170aadefb7afdf3571cf54544561cf63b9f1
3
  size 5905