Hugging Face's logo

Amirhossein75
/
speech-intensity-wav2vec

Automatic Speech Recognition
Transformers
Safetensors
English
wav2vec2
speech
asr
audio-regression
multitask-learning
whisper
gradio
sagemaker
custom_code
Model card Files
xet
Community
Amirhossein75 commited on
Commit
1e3e764
·
1 Parent(s): 7a1ea1d

add new weight and metric to readme file

Browse files
Files changed (4) hide show
  1. README.md +37 -2
  2. config.json +1 -1
  3. model.safetensors +1 -1
  4. training_args.bin +1 -1
README.md CHANGED
@@ -129,9 +129,44 @@ norm_intensity = (dbfs + 60) / 60
129
  - **ASR:** Word Error Rate (WER)
130
  - **Intensity regression:** MAE, MSE, and R²
131
 
132
- <!-- ### Results
133
- _Add your experiment table (WER and regression metrics for Whisper and Wav2Vec2‑CTC)._ -->
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
134
 
 
135
  #### Summary
136
  Multitask objective = ASR loss + intensity regression loss (weight controlled by `--lambda_intensity`).
137
 
 
129
  - **ASR:** Word Error Rate (WER)
130
  - **Intensity regression:** MAE, MSE, and R²
131
 
132
+ ### Results
133
+ ## 📊 Training Logs & Metrics
134
+
135
+ - **Total FLOPs (training):** `11,971,980,681,992,470,000`
136
+ - **Training runtime:** `9,579.8516` seconds for 3 `epoch`
137
+ - **Logging:** TensorBoard-compatible logs in `src/checkpoint/logs`
138
+
139
+ You can monitor training live with:
140
+
141
+ ## ✅ Full Metrics
142
+ ### 🔎 Highlights
143
+ - **Validation WER (↓):** **12.897%** _(0.128966 as fraction)_
144
+ - **Validation Loss:** **21.7842**
145
+ - Fast eval throughput: **17.05 samples/s** • **4.264 steps/s**
146
+
147
+ > **WER** from `jiwer.wer` (fraction in \[0,1\]; percent shown for readability).
148
+ > This run uses a **CTC** objective for ASR and an auxiliary **intensity** head (multi‑task), but only ASR metrics were logged during evaluation.
149
+
150
+ #### Validation (Dev)
151
+ | Metric | Value |
152
+ |---|---|
153
+ | **Loss** | **21.7842** |
154
+ | **WER (↓)** | **0.128966** _(12.897%)_ |
155
+ | **Runtime (s)** | **158.5324** _(≈ 2m 39s)_ |
156
+ | **Samples / s** | **17.050** |
157
+ | **Steps / s** | **4.264** |
158
+ | **Epoch** | **2.8** |
159
+
160
+ #### Training Summary
161
+ | Metric | Value |
162
+ |---|---|
163
+ | **Train Loss** | **227.4951** |
164
+ | **Runtime (s)** | **9,579.8514** _(≈ 2h 39m 40s)_ |
165
+ | **Samples / s** | **8.937** |
166
+ | **Steps / s** | **0.559** |
167
+ | **Epochs** | **3.0** |
168
 
169
+ ---
170
  #### Summary
171
  Multitask objective = ASR loss + intensity regression loss (weight controlled by `--lambda_intensity`).
172
 
config.json CHANGED
@@ -102,7 +102,7 @@
102
  1
103
  ],
104
  "torch_dtype": "float32",
105
- "transformers_version": "4.55.0",
106
  "use_weighted_layer_sum": false,
107
  "vocab_size": 32,
108
  "xvector_output_dim": 512
 
102
  1
103
  ],
104
  "torch_dtype": "float32",
105
+ "transformers_version": "4.55.4",
106
  "use_weighted_layer_sum": false,
107
  "vocab_size": 32,
108
  "xvector_output_dim": 512
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:1052774f090f3e43ce7c9c865dd189fc2c883498e71e50d3ff7a030e95149b1a
3
  size 378005716
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:99692058cd38f6ed6f4a8ca9a62dea8074540aa61be279af535d7d7acddf71e2
3
  size 378005716
training_args.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:190e9f49452411c8be72c56591407cb7f1a25a190f54aefaac34201968d2ffc5
3
  size 5713
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6d7726847e7ccdd715b19f6fae3056bdc8260549f86474484d248113008486b4
3
  size 5713