Spaces:
Paused
Paused
Expose the Hugging Face Code as an API
Browse files
app.py
CHANGED
@@ -1,8 +1,10 @@
|
|
1 |
import os
|
2 |
import torch
|
3 |
-
import gradio as gr
|
4 |
import numpy as np
|
5 |
import soundfile as sf
|
|
|
|
|
|
|
6 |
from transformers import (
|
7 |
AutoModelForSeq2SeqLM,
|
8 |
AutoTokenizer,
|
@@ -13,13 +15,11 @@ from transformers import (
|
|
13 |
WhisperForConditionalGeneration
|
14 |
)
|
15 |
from typing import Optional, Tuple, Dict, List
|
|
|
|
|
16 |
|
|
|
17 |
class TalklasTranslator:
|
18 |
-
"""
|
19 |
-
Speech-to-Speech translation pipeline for Philippine languages.
|
20 |
-
Uses MMS/Whisper for STT, NLLB for MT, and MMS for TTS.
|
21 |
-
"""
|
22 |
-
|
23 |
LANGUAGE_MAPPING = {
|
24 |
"English": "eng",
|
25 |
"Tagalog": "tgl",
|
@@ -50,45 +50,34 @@ class TalklasTranslator:
|
|
50 |
self.sample_rate = 16000
|
51 |
|
52 |
print(f"Initializing Talklas Translator on {self.device}")
|
53 |
-
|
54 |
-
# Initialize models
|
55 |
self._initialize_stt_model()
|
56 |
self._initialize_mt_model()
|
57 |
self._initialize_tts_model()
|
58 |
|
59 |
def _initialize_stt_model(self):
|
60 |
-
"""Initialize speech-to-text model with fallback to Whisper"""
|
61 |
try:
|
62 |
print("Loading STT model...")
|
63 |
try:
|
64 |
-
# Try loading MMS model first
|
65 |
self.stt_processor = AutoProcessor.from_pretrained("facebook/mms-1b-all")
|
66 |
self.stt_model = AutoModelForCTC.from_pretrained("facebook/mms-1b-all")
|
67 |
-
|
68 |
-
# Set language if available
|
69 |
if self.source_lang in self.stt_processor.tokenizer.vocab.keys():
|
70 |
self.stt_processor.tokenizer.set_target_lang(self.source_lang)
|
71 |
self.stt_model.load_adapter(self.source_lang)
|
72 |
print(f"Loaded MMS STT model for {self.source_lang}")
|
73 |
else:
|
74 |
print(f"Language {self.source_lang} not in MMS, using default")
|
75 |
-
|
76 |
except Exception as mms_error:
|
77 |
print(f"MMS loading failed: {mms_error}")
|
78 |
-
# Fallback to Whisper
|
79 |
print("Loading Whisper as fallback...")
|
80 |
self.stt_processor = WhisperProcessor.from_pretrained("openai/whisper-small")
|
81 |
self.stt_model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-small")
|
82 |
print("Loaded Whisper STT model")
|
83 |
-
|
84 |
self.stt_model.to(self.device)
|
85 |
-
|
86 |
except Exception as e:
|
87 |
print(f"STT model initialization failed: {e}")
|
88 |
raise RuntimeError("Could not initialize STT model")
|
89 |
|
90 |
def _initialize_mt_model(self):
|
91 |
-
"""Initialize machine translation model"""
|
92 |
try:
|
93 |
print("Loading NLLB Translation model...")
|
94 |
self.mt_model = AutoModelForSeq2SeqLM.from_pretrained("facebook/nllb-200-distilled-600M")
|
@@ -100,7 +89,6 @@ class TalklasTranslator:
|
|
100 |
raise
|
101 |
|
102 |
def _initialize_tts_model(self):
|
103 |
-
"""Initialize text-to-speech model"""
|
104 |
try:
|
105 |
print("Loading TTS model...")
|
106 |
try:
|
@@ -112,102 +100,78 @@ class TalklasTranslator:
|
|
112 |
print("Falling back to English TTS")
|
113 |
self.tts_model = VitsModel.from_pretrained("facebook/mms-tts-eng")
|
114 |
self.tts_tokenizer = AutoTokenizer.from_pretrained("facebook/mms-tts-eng")
|
115 |
-
|
116 |
self.tts_model.to(self.device)
|
117 |
except Exception as e:
|
118 |
print(f"TTS model initialization failed: {e}")
|
119 |
raise
|
120 |
|
121 |
def update_languages(self, source_lang: str, target_lang: str) -> str:
|
122 |
-
"""Update languages and reinitialize models if needed"""
|
123 |
if source_lang == self.source_lang and target_lang == self.target_lang:
|
124 |
return "Languages already set"
|
125 |
-
|
126 |
self.source_lang = source_lang
|
127 |
self.target_lang = target_lang
|
128 |
-
|
129 |
-
# Only reinitialize models that depend on language
|
130 |
self._initialize_stt_model()
|
131 |
self._initialize_tts_model()
|
132 |
-
|
133 |
return f"Languages updated to {source_lang} → {target_lang}"
|
134 |
|
135 |
def speech_to_text(self, audio_path: str) -> str:
|
136 |
-
"""Convert speech to text using loaded STT model"""
|
137 |
try:
|
138 |
waveform, sample_rate = sf.read(audio_path)
|
139 |
-
|
140 |
if sample_rate != 16000:
|
141 |
import librosa
|
142 |
waveform = librosa.resample(waveform, orig_sr=sample_rate, target_sr=16000)
|
143 |
-
|
144 |
inputs = self.stt_processor(
|
145 |
waveform,
|
146 |
sampling_rate=16000,
|
147 |
return_tensors="pt"
|
148 |
).to(self.device)
|
149 |
-
|
150 |
with torch.no_grad():
|
151 |
-
if isinstance(self.stt_model, WhisperForConditionalGeneration):
|
152 |
generated_ids = self.stt_model.generate(**inputs)
|
153 |
transcription = self.stt_processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
|
154 |
-
else:
|
155 |
logits = self.stt_model(**inputs).logits
|
156 |
predicted_ids = torch.argmax(logits, dim=-1)
|
157 |
transcription = self.stt_processor.batch_decode(predicted_ids)[0]
|
158 |
-
|
159 |
return transcription
|
160 |
-
|
161 |
except Exception as e:
|
162 |
print(f"Speech recognition failed: {e}")
|
163 |
raise RuntimeError("Speech recognition failed")
|
164 |
|
165 |
def translate_text(self, text: str) -> str:
|
166 |
-
"""Translate text using NLLB model"""
|
167 |
try:
|
168 |
source_code = self.NLLB_LANGUAGE_CODES[self.source_lang]
|
169 |
target_code = self.NLLB_LANGUAGE_CODES[self.target_lang]
|
170 |
-
|
171 |
self.mt_tokenizer.src_lang = source_code
|
172 |
inputs = self.mt_tokenizer(text, return_tensors="pt").to(self.device)
|
173 |
-
|
174 |
with torch.no_grad():
|
175 |
generated_tokens = self.mt_model.generate(
|
176 |
**inputs,
|
177 |
forced_bos_token_id=self.mt_tokenizer.convert_tokens_to_ids(target_code),
|
178 |
max_length=448
|
179 |
)
|
180 |
-
|
181 |
return self.mt_tokenizer.batch_decode(generated_tokens, skip_special_tokens=True)[0]
|
182 |
-
|
183 |
except Exception as e:
|
184 |
print(f"Translation failed: {e}")
|
185 |
raise RuntimeError("Text translation failed")
|
186 |
|
187 |
def text_to_speech(self, text: str) -> Tuple[int, np.ndarray]:
|
188 |
-
"""Convert text to speech"""
|
189 |
try:
|
190 |
inputs = self.tts_tokenizer(text, return_tensors="pt").to(self.device)
|
191 |
-
|
192 |
with torch.no_grad():
|
193 |
output = self.tts_model(**inputs)
|
194 |
-
|
195 |
speech = output.waveform.cpu().numpy().squeeze()
|
196 |
speech = (speech * 32767).astype(np.int16)
|
197 |
-
|
198 |
return self.tts_model.config.sampling_rate, speech
|
199 |
-
|
200 |
except Exception as e:
|
201 |
print(f"Speech synthesis failed: {e}")
|
202 |
raise RuntimeError("Speech synthesis failed")
|
203 |
|
204 |
def translate_speech(self, audio_path: str) -> Dict:
|
205 |
-
"""Full speech-to-speech translation"""
|
206 |
try:
|
207 |
source_text = self.speech_to_text(audio_path)
|
208 |
translated_text = self.translate_text(source_text)
|
209 |
sample_rate, audio = self.text_to_speech(translated_text)
|
210 |
-
|
211 |
return {
|
212 |
"source_text": source_text,
|
213 |
"translated_text": translated_text,
|
@@ -223,11 +187,9 @@ class TalklasTranslator:
|
|
223 |
}
|
224 |
|
225 |
def translate_text_only(self, text: str) -> Dict:
|
226 |
-
"""Text-to-speech translation"""
|
227 |
try:
|
228 |
translated_text = self.translate_text(text)
|
229 |
sample_rate, audio = self.text_to_speech(translated_text)
|
230 |
-
|
231 |
return {
|
232 |
"source_text": text,
|
233 |
"translated_text": translated_text,
|
@@ -251,168 +213,88 @@ class TranslatorSingleton:
|
|
251 |
cls._instance = TalklasTranslator()
|
252 |
return cls._instance
|
253 |
|
254 |
-
|
255 |
-
|
256 |
-
|
257 |
-
|
258 |
-
|
259 |
-
|
260 |
-
|
261 |
-
|
262 |
-
|
263 |
-
|
264 |
-
|
265 |
-
|
266 |
-
|
267 |
-
|
268 |
-
|
269 |
-
|
270 |
-
|
271 |
-
|
272 |
-
|
273 |
-
|
274 |
-
|
275 |
-
|
276 |
-
|
277 |
-
|
278 |
-
|
279 |
-
|
280 |
-
|
281 |
-
|
282 |
-
|
283 |
-
|
284 |
-
|
285 |
-
|
286 |
-
|
287 |
-
|
288 |
-
|
289 |
-
|
290 |
-
|
291 |
-
|
292 |
-
|
293 |
-
|
294 |
-
|
295 |
-
|
296 |
-
|
297 |
-
|
298 |
-
|
299 |
-
|
300 |
-
|
301 |
-
|
302 |
-
|
303 |
-
|
304 |
-
|
305 |
-
|
306 |
-
|
307 |
-
|
308 |
-
|
309 |
-
|
310 |
-
|
311 |
-
|
312 |
-
|
313 |
-
|
314 |
-
|
315 |
-
|
316 |
-
|
317 |
-
|
318 |
-
|
319 |
-
|
320 |
-
|
321 |
-
|
322 |
-
|
323 |
-
|
324 |
-
|
325 |
-
|
326 |
-
|
327 |
-
|
328 |
-
|
329 |
-
|
330 |
-
|
331 |
-
|
332 |
-
|
333 |
-
|
334 |
-
|
335 |
-
autoplay=True
|
336 |
-
)
|
337 |
-
|
338 |
-
with gr.TabItem("Text Input"):
|
339 |
-
with gr.Row():
|
340 |
-
with gr.Column():
|
341 |
-
gr.Markdown("### Text Input")
|
342 |
-
text_input = gr.Textbox(
|
343 |
-
label="Enter text to translate",
|
344 |
-
lines=3
|
345 |
-
)
|
346 |
-
text_translate_btn = gr.Button("Translate Text", variant="primary")
|
347 |
-
|
348 |
-
with gr.Column():
|
349 |
-
gr.Markdown("### Output")
|
350 |
-
text_output = gr.Audio(
|
351 |
-
label="Translated Speech",
|
352 |
-
type="numpy",
|
353 |
-
autoplay=True
|
354 |
-
)
|
355 |
-
|
356 |
-
with gr.Row():
|
357 |
-
with gr.Column():
|
358 |
-
source_text = gr.Textbox(label="Source Text")
|
359 |
-
translated_text = gr.Textbox(label="Translated Text")
|
360 |
-
performance_info = gr.Textbox(label="Performance Metrics")
|
361 |
-
|
362 |
-
# Set up events
|
363 |
-
update_btn.click(
|
364 |
-
lambda source_lang, target_lang: TranslatorSingleton.get_instance().update_languages(
|
365 |
-
TalklasTranslator.LANGUAGE_MAPPING[source_lang],
|
366 |
-
TalklasTranslator.LANGUAGE_MAPPING[target_lang]
|
367 |
-
),
|
368 |
-
inputs=[source_lang, target_lang],
|
369 |
-
outputs=[language_status]
|
370 |
-
)
|
371 |
-
|
372 |
-
# Audio translate button click
|
373 |
-
audio_translate_btn.click(
|
374 |
-
process_audio,
|
375 |
-
inputs=[audio_input, source_lang, target_lang],
|
376 |
-
outputs=[audio_output, source_text, translated_text, performance_info]
|
377 |
-
).then(
|
378 |
-
None,
|
379 |
-
None,
|
380 |
-
None,
|
381 |
-
js="""() => {
|
382 |
-
const audioElements = document.querySelectorAll('audio');
|
383 |
-
if (audioElements.length > 0) {
|
384 |
-
const lastAudio = audioElements[audioElements.length - 1];
|
385 |
-
lastAudio.play().catch(error => {
|
386 |
-
console.warn('Autoplay failed:', error);
|
387 |
-
alert('Audio may require user interaction to play');
|
388 |
-
});
|
389 |
-
}
|
390 |
-
}"""
|
391 |
-
)
|
392 |
-
|
393 |
-
# Text translate button click
|
394 |
-
text_translate_btn.click(
|
395 |
-
process_text,
|
396 |
-
inputs=[text_input, source_lang, target_lang],
|
397 |
-
outputs=[text_output, source_text, translated_text, performance_info]
|
398 |
-
).then(
|
399 |
-
None,
|
400 |
-
None,
|
401 |
-
None,
|
402 |
-
js="""() => {
|
403 |
-
const audioElements = document.querySelectorAll('audio');
|
404 |
-
if (audioElements.length > 0) {
|
405 |
-
const lastAudio = audioElements[audioElements.length - 1];
|
406 |
-
lastAudio.play().catch(error => {
|
407 |
-
console.warn('Autoplay failed:', error);
|
408 |
-
alert('Audio may require user interaction to play');
|
409 |
-
});
|
410 |
-
}
|
411 |
-
}"""
|
412 |
-
)
|
413 |
-
|
414 |
-
return demo
|
415 |
|
416 |
if __name__ == "__main__":
|
417 |
-
|
418 |
-
|
|
|
1 |
import os
|
2 |
import torch
|
|
|
3 |
import numpy as np
|
4 |
import soundfile as sf
|
5 |
+
from fastapi import FastAPI, File, UploadFile, HTTPException
|
6 |
+
from fastapi.responses import JSONResponse
|
7 |
+
from pydantic import BaseModel
|
8 |
from transformers import (
|
9 |
AutoModelForSeq2SeqLM,
|
10 |
AutoTokenizer,
|
|
|
15 |
WhisperForConditionalGeneration
|
16 |
)
|
17 |
from typing import Optional, Tuple, Dict, List
|
18 |
+
import base64
|
19 |
+
import io
|
20 |
|
21 |
+
# Your existing TalklasTranslator class (unchanged)
|
22 |
class TalklasTranslator:
|
|
|
|
|
|
|
|
|
|
|
23 |
LANGUAGE_MAPPING = {
|
24 |
"English": "eng",
|
25 |
"Tagalog": "tgl",
|
|
|
50 |
self.sample_rate = 16000
|
51 |
|
52 |
print(f"Initializing Talklas Translator on {self.device}")
|
|
|
|
|
53 |
self._initialize_stt_model()
|
54 |
self._initialize_mt_model()
|
55 |
self._initialize_tts_model()
|
56 |
|
57 |
def _initialize_stt_model(self):
|
|
|
58 |
try:
|
59 |
print("Loading STT model...")
|
60 |
try:
|
|
|
61 |
self.stt_processor = AutoProcessor.from_pretrained("facebook/mms-1b-all")
|
62 |
self.stt_model = AutoModelForCTC.from_pretrained("facebook/mms-1b-all")
|
|
|
|
|
63 |
if self.source_lang in self.stt_processor.tokenizer.vocab.keys():
|
64 |
self.stt_processor.tokenizer.set_target_lang(self.source_lang)
|
65 |
self.stt_model.load_adapter(self.source_lang)
|
66 |
print(f"Loaded MMS STT model for {self.source_lang}")
|
67 |
else:
|
68 |
print(f"Language {self.source_lang} not in MMS, using default")
|
|
|
69 |
except Exception as mms_error:
|
70 |
print(f"MMS loading failed: {mms_error}")
|
|
|
71 |
print("Loading Whisper as fallback...")
|
72 |
self.stt_processor = WhisperProcessor.from_pretrained("openai/whisper-small")
|
73 |
self.stt_model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-small")
|
74 |
print("Loaded Whisper STT model")
|
|
|
75 |
self.stt_model.to(self.device)
|
|
|
76 |
except Exception as e:
|
77 |
print(f"STT model initialization failed: {e}")
|
78 |
raise RuntimeError("Could not initialize STT model")
|
79 |
|
80 |
def _initialize_mt_model(self):
|
|
|
81 |
try:
|
82 |
print("Loading NLLB Translation model...")
|
83 |
self.mt_model = AutoModelForSeq2SeqLM.from_pretrained("facebook/nllb-200-distilled-600M")
|
|
|
89 |
raise
|
90 |
|
91 |
def _initialize_tts_model(self):
|
|
|
92 |
try:
|
93 |
print("Loading TTS model...")
|
94 |
try:
|
|
|
100 |
print("Falling back to English TTS")
|
101 |
self.tts_model = VitsModel.from_pretrained("facebook/mms-tts-eng")
|
102 |
self.tts_tokenizer = AutoTokenizer.from_pretrained("facebook/mms-tts-eng")
|
|
|
103 |
self.tts_model.to(self.device)
|
104 |
except Exception as e:
|
105 |
print(f"TTS model initialization failed: {e}")
|
106 |
raise
|
107 |
|
108 |
def update_languages(self, source_lang: str, target_lang: str) -> str:
|
|
|
109 |
if source_lang == self.source_lang and target_lang == self.target_lang:
|
110 |
return "Languages already set"
|
|
|
111 |
self.source_lang = source_lang
|
112 |
self.target_lang = target_lang
|
|
|
|
|
113 |
self._initialize_stt_model()
|
114 |
self._initialize_tts_model()
|
|
|
115 |
return f"Languages updated to {source_lang} → {target_lang}"
|
116 |
|
117 |
def speech_to_text(self, audio_path: str) -> str:
|
|
|
118 |
try:
|
119 |
waveform, sample_rate = sf.read(audio_path)
|
|
|
120 |
if sample_rate != 16000:
|
121 |
import librosa
|
122 |
waveform = librosa.resample(waveform, orig_sr=sample_rate, target_sr=16000)
|
|
|
123 |
inputs = self.stt_processor(
|
124 |
waveform,
|
125 |
sampling_rate=16000,
|
126 |
return_tensors="pt"
|
127 |
).to(self.device)
|
|
|
128 |
with torch.no_grad():
|
129 |
+
if isinstance(self.stt_model, WhisperForConditionalGeneration):
|
130 |
generated_ids = self.stt_model.generate(**inputs)
|
131 |
transcription = self.stt_processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
|
132 |
+
else:
|
133 |
logits = self.stt_model(**inputs).logits
|
134 |
predicted_ids = torch.argmax(logits, dim=-1)
|
135 |
transcription = self.stt_processor.batch_decode(predicted_ids)[0]
|
|
|
136 |
return transcription
|
|
|
137 |
except Exception as e:
|
138 |
print(f"Speech recognition failed: {e}")
|
139 |
raise RuntimeError("Speech recognition failed")
|
140 |
|
141 |
def translate_text(self, text: str) -> str:
|
|
|
142 |
try:
|
143 |
source_code = self.NLLB_LANGUAGE_CODES[self.source_lang]
|
144 |
target_code = self.NLLB_LANGUAGE_CODES[self.target_lang]
|
|
|
145 |
self.mt_tokenizer.src_lang = source_code
|
146 |
inputs = self.mt_tokenizer(text, return_tensors="pt").to(self.device)
|
|
|
147 |
with torch.no_grad():
|
148 |
generated_tokens = self.mt_model.generate(
|
149 |
**inputs,
|
150 |
forced_bos_token_id=self.mt_tokenizer.convert_tokens_to_ids(target_code),
|
151 |
max_length=448
|
152 |
)
|
|
|
153 |
return self.mt_tokenizer.batch_decode(generated_tokens, skip_special_tokens=True)[0]
|
|
|
154 |
except Exception as e:
|
155 |
print(f"Translation failed: {e}")
|
156 |
raise RuntimeError("Text translation failed")
|
157 |
|
158 |
def text_to_speech(self, text: str) -> Tuple[int, np.ndarray]:
|
|
|
159 |
try:
|
160 |
inputs = self.tts_tokenizer(text, return_tensors="pt").to(self.device)
|
|
|
161 |
with torch.no_grad():
|
162 |
output = self.tts_model(**inputs)
|
|
|
163 |
speech = output.waveform.cpu().numpy().squeeze()
|
164 |
speech = (speech * 32767).astype(np.int16)
|
|
|
165 |
return self.tts_model.config.sampling_rate, speech
|
|
|
166 |
except Exception as e:
|
167 |
print(f"Speech synthesis failed: {e}")
|
168 |
raise RuntimeError("Speech synthesis failed")
|
169 |
|
170 |
def translate_speech(self, audio_path: str) -> Dict:
|
|
|
171 |
try:
|
172 |
source_text = self.speech_to_text(audio_path)
|
173 |
translated_text = self.translate_text(source_text)
|
174 |
sample_rate, audio = self.text_to_speech(translated_text)
|
|
|
175 |
return {
|
176 |
"source_text": source_text,
|
177 |
"translated_text": translated_text,
|
|
|
187 |
}
|
188 |
|
189 |
def translate_text_only(self, text: str) -> Dict:
|
|
|
190 |
try:
|
191 |
translated_text = self.translate_text(text)
|
192 |
sample_rate, audio = self.text_to_speech(translated_text)
|
|
|
193 |
return {
|
194 |
"source_text": text,
|
195 |
"translated_text": translated_text,
|
|
|
213 |
cls._instance = TalklasTranslator()
|
214 |
return cls._instance
|
215 |
|
216 |
+
# FastAPI application
|
217 |
+
app = FastAPI(title="Talklas API", description="Speech-to-Speech Translation API")
|
218 |
+
|
219 |
+
class TranslationRequest(BaseModel):
|
220 |
+
source_lang: str
|
221 |
+
target_lang: str
|
222 |
+
text: Optional[str] = None
|
223 |
+
|
224 |
+
@app.post("/translate/audio")
|
225 |
+
async def translate_audio(file: UploadFile = File(...), source_lang: str = "English", target_lang: str = "Tagalog"):
|
226 |
+
try:
|
227 |
+
# Validate languages
|
228 |
+
if source_lang not in TalklasTranslator.LANGUAGE_MAPPING or target_lang not in TalklasTranslator.LANGUAGE_MAPPING:
|
229 |
+
raise HTTPException(status_code=400, detail="Invalid language selection")
|
230 |
+
|
231 |
+
# Save uploaded audio file temporarily
|
232 |
+
audio_path = f"temp_{file.filename}"
|
233 |
+
with open(audio_path, "wb") as f:
|
234 |
+
f.write(await file.read())
|
235 |
+
|
236 |
+
# Update languages
|
237 |
+
source_code = TalklasTranslator.LANGUAGE_MAPPING[source_lang]
|
238 |
+
target_code = TalklasTranslator.LANGUAGE_MAPPING[target_lang]
|
239 |
+
translator = TranslatorSingleton.get_instance()
|
240 |
+
translator.update_languages(source_code, target_code)
|
241 |
+
|
242 |
+
# Process the audio
|
243 |
+
results = translator.translate_speech(audio_path)
|
244 |
+
|
245 |
+
# Clean up temporary file
|
246 |
+
os.remove(audio_path)
|
247 |
+
|
248 |
+
# Convert audio to base64 for response
|
249 |
+
sample_rate, audio = results["output_audio"]
|
250 |
+
buffer = io.BytesIO()
|
251 |
+
sf.write(buffer, audio, sample_rate, format="wav")
|
252 |
+
audio_base64 = base64.b64encode(buffer.getvalue()).decode("utf-8")
|
253 |
+
|
254 |
+
return JSONResponse(content={
|
255 |
+
"source_text": results["source_text"],
|
256 |
+
"translated_text": results["translated_text"],
|
257 |
+
"audio_base64": audio_base64,
|
258 |
+
"sample_rate": sample_rate,
|
259 |
+
"performance": results["performance"]
|
260 |
+
})
|
261 |
+
except Exception as e:
|
262 |
+
raise HTTPException(status_code=500, detail=f"Translation failed: {str(e)}")
|
263 |
+
|
264 |
+
@app.post("/translate/text")
|
265 |
+
async def translate_text(request: TranslationRequest):
|
266 |
+
try:
|
267 |
+
# Validate input
|
268 |
+
if not request.text:
|
269 |
+
raise HTTPException(status_code=400, detail="Text input is required")
|
270 |
+
if request.source_lang not in TalklasTranslator.LANGUAGE_MAPPING or request.target_lang not in TalklasTranslator.LANGUAGE_MAPPING:
|
271 |
+
raise HTTPException(status_code=400, detail="Invalid language selection")
|
272 |
+
|
273 |
+
# Update languages
|
274 |
+
source_code = TalklasTranslator.LANGUAGE_MAPPING[request.source_lang]
|
275 |
+
target_code = TalklasTranslator.LANGUAGE_MAPPING[request.target_lang]
|
276 |
+
translator = TranslatorSingleton.get_instance()
|
277 |
+
translator.update_languages(source_code, target_code)
|
278 |
+
|
279 |
+
# Process the text
|
280 |
+
results = translator.translate_text_only(request.text)
|
281 |
+
|
282 |
+
# Convert audio to base64 for response
|
283 |
+
sample_rate, audio = results["output_audio"]
|
284 |
+
buffer = io.BytesIO()
|
285 |
+
sf.write(buffer, audio, sample_rate, format="wav")
|
286 |
+
audio_base64 = base64.b64encode(buffer.getvalue()).decode("utf-8")
|
287 |
+
|
288 |
+
return JSONResponse(content={
|
289 |
+
"source_text": results["source_text"],
|
290 |
+
"translated_text": results["translated_text"],
|
291 |
+
"audio_base64": audio_base64,
|
292 |
+
"sample_rate": sample_rate,
|
293 |
+
"performance": results["performance"]
|
294 |
+
})
|
295 |
+
except Exception as e:
|
296 |
+
raise HTTPException(status_code=500, detail=f"Translation failed: {str(e)}")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
297 |
|
298 |
if __name__ == "__main__":
|
299 |
+
import uvicorn
|
300 |
+
uvicorn.run(app, host="0.0.0.0", port=8000)
|