Spaces:
Running
Running
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,116 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
import torchaudio
|
3 |
+
import scipy.io.wavfile
|
4 |
+
from transformers import AutoProcessor, SeamlessM4Tv2Model
|
5 |
+
from pathlib import Path
|
6 |
+
from typing import Optional, Union
|
7 |
+
|
8 |
+
class SeamlessTranslator:
|
9 |
+
"""
|
10 |
+
A wrapper class for Facebook's SeamlessM4T translation model.
|
11 |
+
Handles both text-to-speech and speech-to-speech translation.
|
12 |
+
"""
|
13 |
+
|
14 |
+
def __init__(self, model_name: str = "facebook/seamless-m4t-v2-large"):
|
15 |
+
"""
|
16 |
+
Initialize the translator with the specified model.
|
17 |
+
|
18 |
+
Args:
|
19 |
+
model_name (str): Name of the model to use
|
20 |
+
"""
|
21 |
+
try:
|
22 |
+
self.processor = AutoProcessor.from_pretrained(model_name)
|
23 |
+
self.model = SeamlessM4Tv2Model.from_pretrained(model_name)
|
24 |
+
self.sample_rate = self.model.config.sampling_rate
|
25 |
+
except Exception as e:
|
26 |
+
raise RuntimeError(f"Failed to initialize model: {str(e)}")
|
27 |
+
|
28 |
+
def translate_text(self, text: str, src_lang: str, tgt_lang: str) -> numpy.ndarray:
|
29 |
+
"""
|
30 |
+
Translate text to speech in the target language.
|
31 |
+
|
32 |
+
Args:
|
33 |
+
text (str): Input text to translate
|
34 |
+
src_lang (str): Source language code (e.g., 'eng')
|
35 |
+
tgt_lang (str): Target language code (e.g., 'rus')
|
36 |
+
|
37 |
+
Returns:
|
38 |
+
numpy.ndarray: Audio waveform array
|
39 |
+
"""
|
40 |
+
try:
|
41 |
+
inputs = self.processor(text=text, src_lang=src_lang, return_tensors="pt")
|
42 |
+
audio_array = self.model.generate(**inputs, tgt_lang=tgt_lang)[0].cpu().numpy().squeeze()
|
43 |
+
return audio_array
|
44 |
+
except Exception as e:
|
45 |
+
raise RuntimeError(f"Text translation failed: {str(e)}")
|
46 |
+
|
47 |
+
def translate_audio(self, audio_path: Union[str, Path], tgt_lang: str) -> numpy.ndarray:
|
48 |
+
"""
|
49 |
+
Translate audio to speech in the target language.
|
50 |
+
|
51 |
+
Args:
|
52 |
+
audio_path (str or Path): Path to input audio file
|
53 |
+
tgt_lang (str): Target language code (e.g., 'rus')
|
54 |
+
|
55 |
+
Returns:
|
56 |
+
numpy.ndarray: Audio waveform array
|
57 |
+
"""
|
58 |
+
try:
|
59 |
+
# Load and resample audio
|
60 |
+
audio, orig_freq = torchaudio.load(audio_path)
|
61 |
+
audio = torchaudio.functional.resample(
|
62 |
+
audio,
|
63 |
+
orig_freq=orig_freq,
|
64 |
+
new_freq=16_000
|
65 |
+
)
|
66 |
+
|
67 |
+
# Process and generate translation
|
68 |
+
inputs = self.processor(audios=audio, return_tensors="pt")
|
69 |
+
audio_array = self.model.generate(**inputs, tgt_lang=tgt_lang)[0].cpu().numpy().squeeze()
|
70 |
+
return audio_array
|
71 |
+
except Exception as e:
|
72 |
+
raise RuntimeError(f"Audio translation failed: {str(e)}")
|
73 |
+
|
74 |
+
def save_audio(self, audio_array: numpy.ndarray, output_path: Union[str, Path]) -> None:
|
75 |
+
"""
|
76 |
+
Save an audio array to a WAV file.
|
77 |
+
|
78 |
+
Args:
|
79 |
+
audio_array (numpy.ndarray): Audio data to save
|
80 |
+
output_path (str or Path): Path where to save the WAV file
|
81 |
+
"""
|
82 |
+
try:
|
83 |
+
scipy.io.wavfile.write(
|
84 |
+
output_path,
|
85 |
+
rate=self.sample_rate,
|
86 |
+
data=audio_array
|
87 |
+
)
|
88 |
+
except Exception as e:
|
89 |
+
raise RuntimeError(f"Failed to save audio: {str(e)}")
|
90 |
+
|
91 |
+
def main():
|
92 |
+
"""Example usage of the SeamlessTranslator class."""
|
93 |
+
try:
|
94 |
+
# Initialize translator
|
95 |
+
translator = SeamlessTranslator()
|
96 |
+
|
97 |
+
# Example text translation
|
98 |
+
text_audio = translator.translate_text(
|
99 |
+
text="Hello, my dog is cute",
|
100 |
+
src_lang="eng",
|
101 |
+
tgt_lang="rus"
|
102 |
+
)
|
103 |
+
translator.save_audio(text_audio, "output_from_text.wav")
|
104 |
+
|
105 |
+
# Example audio translation
|
106 |
+
audio_audio = translator.translate_audio(
|
107 |
+
audio_path="input_audio.wav",
|
108 |
+
tgt_lang="rus"
|
109 |
+
)
|
110 |
+
translator.save_audio(audio_audio, "output_from_audio.wav")
|
111 |
+
|
112 |
+
except Exception as e:
|
113 |
+
print(f"Translation failed: {str(e)}")
|
114 |
+
|
115 |
+
if __name__ == "__main__":
|
116 |
+
main()
|