Update README.md
Browse files
README.md
CHANGED
@@ -18,6 +18,40 @@ grammatical mistakes, slang, and non-native speaker errors. This model helps imp
|
|
18 |
in scenarios where speakers use incorrect or informal English, making it useful in language learning,
|
19 |
transcription of casual conversations, or analyzing spoken communication from non-native English speakers.
|
20 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
21 |
## Usage Guide
|
22 |
|
23 |
This project was executed on an Ubuntu 22.04.3 system running Linux kernel 6.8.0-40-generic.
|
@@ -80,4 +114,12 @@ model.generation_config.task = "transcribe"
|
|
80 |
tokenizer = WhisperTokenizer.from_pretrained("openai/whisper-large-v3", task="transcribe")
|
81 |
feature_extractor = WhisperFeatureExtractor.from_pretrained("openai/whisper-large-v3")
|
82 |
|
83 |
-
pipe = pipeline(model=model, tokenizer=tokenizer, feature_extractor=feature_extractor, task="automatic-speech-recognition", device=device)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
18 |
in scenarios where speakers use incorrect or informal English, making it useful in language learning,
|
19 |
transcription of casual conversations, or analyzing spoken communication from non-native English speakers.
|
20 |
|
21 |
+
## Training procedure
|
22 |
+
|
23 |
+
### Training hyperparameters
|
24 |
+
|
25 |
+
The following hyperparameters were used during training:
|
26 |
+
- learning_rate: 1e-05
|
27 |
+
- train_batch_size: 28
|
28 |
+
- eval_batch_size: 28
|
29 |
+
- seed: 42
|
30 |
+
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
31 |
+
- lr_scheduler_type: linear
|
32 |
+
- lr_scheduler_warmup_steps: 50
|
33 |
+
- training_steps: 100000
|
34 |
+
- mixed_precision_training: Native AMP
|
35 |
+
|
36 |
+
### Training results
|
37 |
+
|
38 |
+
| Training Loss | Epoch | Step | Validation Loss | Wer |
|
39 |
+
|:-------------:|:------:|:----:|:---------------:|:-------:|
|
40 |
+
| 1.5189 | 0.4444 | 500 | 1.1913 | 25.9108 |
|
41 |
+
| 1.1727 | 0.8889 | 1000 | 0.9531 | 24.5396 |
|
42 |
+
| 1.1341 | 1.3333 | 1500 | 0.8688 | 22.2761 |
|
43 |
+
| 1.0152 | 1.7778 | 2000 | 0.8174 | 20.8792 |
|
44 |
+
| 1.0589 | 2.2222 | 2500 | 0.7855 | 20.7595 |
|
45 |
+
| 0.9793 | 2.6667 | 3000 | 0.7611 | 22.2846 |
|
46 |
+
| 0.9594 | 3.1111 | 3500 | 0.7442 | 20.3860 |
|
47 |
+
| 1.0031 | 3.5556 | 4000 | 0.7303 | 18.5045 |
|
48 |
+
| 0.9525 | 4.0 | 4500 | 0.7199 | 18.1054 |
|
49 |
+
| 0.8729 | 4.4444 | 5000 | 0.7105 | 19.3170 |
|
50 |
+
| 1.0031 | 4.8889 | 5500 | 0.7028 | 19.7446 |
|
51 |
+
| 0.9273 | 5.3333 | 6000 | 0.6966 | 19.7189 |
|
52 |
+
| 0.9174 | 5.7778 | 6500 | 0.6896 | 18.4475 |
|
53 |
+
| 0.8842 | 6.2222 | 7000 | 0.6839 | 18.4361 |
|
54 |
+
|
55 |
## Usage Guide
|
56 |
|
57 |
This project was executed on an Ubuntu 22.04.3 system running Linux kernel 6.8.0-40-generic.
|
|
|
114 |
tokenizer = WhisperTokenizer.from_pretrained("openai/whisper-large-v3", task="transcribe")
|
115 |
feature_extractor = WhisperFeatureExtractor.from_pretrained("openai/whisper-large-v3")
|
116 |
|
117 |
+
pipe = pipeline(model=model, tokenizer=tokenizer, feature_extractor=feature_extractor, task="automatic-speech-recognition", device=device)
|
118 |
+
|
119 |
+
### Framework versions
|
120 |
+
|
121 |
+
- PEFT 0.11.1
|
122 |
+
- Transformers 4.42.4
|
123 |
+
- Pytorch 2.1.0+cu118
|
124 |
+
- Datasets 2.20.0
|
125 |
+
- Tokenizers 0.19.1
|