Upload STT Training Script.py
Browse files- STT Training Script.py +216 -0
STT Training Script.py
ADDED
@@ -0,0 +1,216 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Import required libraries
|
2 |
+
from datasets import load_dataset, Audio
|
3 |
+
from transformers import (
|
4 |
+
WhisperProcessor,
|
5 |
+
WhisperForConditionalGeneration,
|
6 |
+
Seq2SeqTrainingArguments,
|
7 |
+
Seq2SeqTrainer
|
8 |
+
)
|
9 |
+
import torch
|
10 |
+
from dataclasses import dataclass
|
11 |
+
from typing import Any, Dict, List, Union
|
12 |
+
from functools import partial
|
13 |
+
import evaluate
|
14 |
+
|
15 |
+
# Load the dataset
|
16 |
+
dataset = load_dataset("") # Specify Data Repo on HF
|
17 |
+
dataset
|
18 |
+
|
19 |
+
# Split the dataset into train and test sets (80-20 split)
|
20 |
+
split_dataset = dataset['train'].train_test_split(test_size=0.2)
|
21 |
+
split_dataset
|
22 |
+
|
23 |
+
# Select only the relevant columns for training
|
24 |
+
split_dataset['train'] = split_dataset['train'].select_columns(["audio", "sentence"])
|
25 |
+
split_dataset['train']
|
26 |
+
|
27 |
+
# Initialize the Whisper processor for Swahili transcription
|
28 |
+
processor = WhisperProcessor.from_pretrained(
|
29 |
+
"openai/whisper-small",
|
30 |
+
language="swahili",
|
31 |
+
task="transcribe"
|
32 |
+
)
|
33 |
+
|
34 |
+
# Print audio features before and after resampling to match Whisper's expected sampling rate
|
35 |
+
print('BEFORE>>> ', split_dataset['train'].features['audio'])
|
36 |
+
sampling_rate = processor.feature_extractor.sampling_rate
|
37 |
+
split_dataset['train'] = split_dataset['train'].cast_column(
|
38 |
+
"audio",
|
39 |
+
Audio(sampling_rate=sampling_rate)
|
40 |
+
)
|
41 |
+
print('AFTER>>> ', split_dataset['train'].features['audio'])
|
42 |
+
|
43 |
+
# Do the same for the test set
|
44 |
+
print('BEFORE>>> ', split_dataset['test'].features['audio'])
|
45 |
+
split_dataset['test'] = split_dataset['test'].cast_column(
|
46 |
+
"audio",
|
47 |
+
Audio(sampling_rate=sampling_rate)
|
48 |
+
)
|
49 |
+
print('AFTER>>> ', split_dataset['test'].features['audio'])
|
50 |
+
|
51 |
+
def prepare_dataset(example):
|
52 |
+
"""Preprocess audio and text data for Whisper model training"""
|
53 |
+
audio = example["audio"]
|
54 |
+
|
55 |
+
# Process audio and text using Whisper processor
|
56 |
+
example = processor(
|
57 |
+
audio=audio["array"],
|
58 |
+
sampling_rate=audio["sampling_rate"],
|
59 |
+
text=example["sentence"],
|
60 |
+
)
|
61 |
+
|
62 |
+
# Compute input length of audio sample in seconds
|
63 |
+
example["input_length"] = len(audio["array"]) / audio["sampling_rate"]
|
64 |
+
|
65 |
+
return example
|
66 |
+
|
67 |
+
# Apply preprocessing to train and test sets
|
68 |
+
split_dataset['train'] = split_dataset['train'].map(
|
69 |
+
prepare_dataset,
|
70 |
+
remove_columns=split_dataset['train'].column_names,
|
71 |
+
num_proc=4 # Use 4 processes for faster preprocessing
|
72 |
+
)
|
73 |
+
|
74 |
+
split_dataset['test'] = split_dataset['test'].map(
|
75 |
+
prepare_dataset,
|
76 |
+
remove_columns=split_dataset['test'].column_names,
|
77 |
+
num_proc=1
|
78 |
+
)
|
79 |
+
|
80 |
+
# Filter out audio samples longer than 30 seconds
|
81 |
+
max_input_length = 30.0
|
82 |
+
def is_audio_in_length_range(length):
|
83 |
+
return length < max_input_length
|
84 |
+
|
85 |
+
split_dataset['train'] = split_dataset['train'].filter(
|
86 |
+
is_audio_in_length_range,
|
87 |
+
input_columns=["input_length"],
|
88 |
+
)
|
89 |
+
|
90 |
+
@dataclass
|
91 |
+
class DataCollatorSpeechSeq2SeqWithPadding:
|
92 |
+
"""Custom data collator for Whisper speech-to-sequence tasks with padding"""
|
93 |
+
processor: Any
|
94 |
+
|
95 |
+
def __call__(
|
96 |
+
self, features: List[Dict[str, Union[List[int], torch.Tensor]]]
|
97 |
+
) -> Dict[str, torch.Tensor]:
|
98 |
+
# Split inputs and labels since they need different padding methods
|
99 |
+
# First process audio inputs
|
100 |
+
input_features = [
|
101 |
+
{"input_features": feature["input_features"][0]} for feature in features
|
102 |
+
]
|
103 |
+
batch = self.processor.feature_extractor.pad(input_features, return_tensors="pt")
|
104 |
+
|
105 |
+
# Process label sequences
|
106 |
+
label_features = [{"input_ids": feature["labels"]} for feature in features]
|
107 |
+
labels_batch = self.processor.tokenizer.pad(label_features, return_tensors="pt")
|
108 |
+
|
109 |
+
# Replace padding with -100 to ignore loss correctly
|
110 |
+
labels = labels_batch["input_ids"].masked_fill(
|
111 |
+
labels_batch.attention_mask.ne(1), -100
|
112 |
+
)
|
113 |
+
|
114 |
+
# Remove BOS token if it was appended previously
|
115 |
+
if (labels[:, 0] == self.processor.tokenizer.bos_token_id).all().cpu().item():
|
116 |
+
labels = labels[:, 1:]
|
117 |
+
|
118 |
+
batch["labels"] = labels
|
119 |
+
|
120 |
+
return batch
|
121 |
+
|
122 |
+
# Initialize data collator
|
123 |
+
data_collator = DataCollatorSpeechSeq2SeqWithPadding(processor=processor)
|
124 |
+
|
125 |
+
# Load evaluation metric (Word Error Rate)
|
126 |
+
metric = evaluate.load("wer")
|
127 |
+
|
128 |
+
# Initialize text normalizer for evaluation
|
129 |
+
from transformers.models.whisper.english_normalizer import BasicTextNormalizer
|
130 |
+
normalizer = BasicTextNormalizer()
|
131 |
+
|
132 |
+
def compute_metrics(pred):
|
133 |
+
"""Compute WER (Word Error Rate) metrics for evaluation"""
|
134 |
+
pred_ids = pred.predictions
|
135 |
+
label_ids = pred.label_ids
|
136 |
+
|
137 |
+
# Replace -100 with pad_token_id
|
138 |
+
label_ids[label_ids == -100] = processor.tokenizer.pad_token_id
|
139 |
+
|
140 |
+
# Decode predictions and labels
|
141 |
+
pred_str = processor.batch_decode(pred_ids, skip_special_tokens=True)
|
142 |
+
label_str = processor.batch_decode(label_ids, skip_special_tokens=True)
|
143 |
+
|
144 |
+
# Compute orthographic WER
|
145 |
+
wer_ortho = 100 * metric.compute(predictions=pred_str, references=label_str)
|
146 |
+
|
147 |
+
# Compute normalized WER
|
148 |
+
pred_str_norm = [normalizer(pred) for pred in pred_str]
|
149 |
+
label_str_norm = [normalizer(label) for label in label_str]
|
150 |
+
|
151 |
+
# Filter samples with non-zero references
|
152 |
+
pred_str_norm = [
|
153 |
+
pred_str_norm[i] for i in range(len(pred_str_norm)) if len(label_str_norm[i]) > 0
|
154 |
+
]
|
155 |
+
label_str_norm = [
|
156 |
+
label_str_norm[i] for i in range(len(label_str_norm)) if len(label_str_norm[i]) > 0
|
157 |
+
]
|
158 |
+
|
159 |
+
wer = 100 * metric.compute(predictions=pred_str_norm, references=label_str_norm)
|
160 |
+
|
161 |
+
return {"wer_ortho": wer_ortho, "wer": wer}
|
162 |
+
|
163 |
+
# Load pre-trained Whisper model
|
164 |
+
model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-small")
|
165 |
+
|
166 |
+
# Disable cache during training (incompatible with gradient checkpointing)
|
167 |
+
model.config.use_cache = False
|
168 |
+
|
169 |
+
# Configure generation settings (re-enable cache for generation)
|
170 |
+
model.generate = partial(
|
171 |
+
model.generate,
|
172 |
+
language="swahili",
|
173 |
+
task="transcribe",
|
174 |
+
use_cache=True
|
175 |
+
)
|
176 |
+
|
177 |
+
# Set up training arguments
|
178 |
+
training_args = Seq2SeqTrainingArguments(
|
179 |
+
output_dir="./model", # Output directory
|
180 |
+
per_device_train_batch_size=16, # Batch size for training
|
181 |
+
gradient_accumulation_steps=1, # Number of steps before gradient update
|
182 |
+
learning_rate=1e-6, # Learning rate
|
183 |
+
lr_scheduler_type="constant_with_warmup", # Learning rate scheduler
|
184 |
+
warmup_steps=50, # Warmup steps
|
185 |
+
max_steps=10000, # Total training steps
|
186 |
+
gradient_checkpointing=True, # Use gradient checkpointing
|
187 |
+
fp16=True, # Use mixed precision training
|
188 |
+
fp16_full_eval=True, # Use mixed precision evaluation
|
189 |
+
evaluation_strategy="steps", # Evaluation strategy
|
190 |
+
per_device_eval_batch_size=16, # Batch size for evaluation
|
191 |
+
predict_with_generate=True, # Use generation for evaluation
|
192 |
+
generation_max_length=225, # Maximum generation length
|
193 |
+
save_steps=500, # Save checkpoint every N steps
|
194 |
+
eval_steps=500, # Evaluate every N steps
|
195 |
+
logging_steps=100, # Log metrics every N steps
|
196 |
+
report_to=["tensorboard", "wandb"], # Logging integrations
|
197 |
+
load_best_model_at_end=True, # Load best model at end of training
|
198 |
+
metric_for_best_model="wer", # Metric for selecting best model
|
199 |
+
greater_is_better=False, # Lower WER is better
|
200 |
+
push_to_hub=True, # Push to Hugging Face Hub
|
201 |
+
save_total_limit=3, # Maximum number of checkpoints to keep
|
202 |
+
)
|
203 |
+
|
204 |
+
# Initialize trainer
|
205 |
+
trainer = Seq2SeqTrainer(
|
206 |
+
args=training_args,
|
207 |
+
model=model,
|
208 |
+
train_dataset=split_dataset['train'],
|
209 |
+
eval_dataset=split_dataset['test'],
|
210 |
+
data_collator=data_collator,
|
211 |
+
compute_metrics=compute_metrics,
|
212 |
+
tokenizer=processor, # Changed from processing_class to tokenizer
|
213 |
+
)
|
214 |
+
|
215 |
+
# Start training
|
216 |
+
trainer.train()
|