Spaces:
Running
Running
Create speech_api.py
Browse files- speech_api.py +65 -0
speech_api.py
ADDED
@@ -0,0 +1,65 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from fastapi import APIRouter, File, UploadFile, HTTPException
|
2 |
+
from fastapi.responses import StreamingResponse
|
3 |
+
from pydantic import BaseModel
|
4 |
+
from ai4b import BhashiniClient
|
5 |
+
from fast_langdetect import detect
|
6 |
+
import io
|
7 |
+
import base64
|
8 |
+
|
9 |
+
router = APIRouter()
|
10 |
+
|
11 |
+
ULCA_USER_ID = os.getenv("ULCA_USER_ID")
|
12 |
+
ULCA_API_KEY = os.geteenv("ULCA_API_KEY")
|
13 |
+
|
14 |
+
client = BhashiniClient(user_id=USER_ID, api_key=ULCA_API_KEY)
|
15 |
+
|
16 |
+
class TTSRequest(BaseModel):
|
17 |
+
text: str
|
18 |
+
gender: str = "female"
|
19 |
+
|
20 |
+
SUPPORTED_LANGUAGES = {'pa', 'mr', 'bn', 'en', 'as', 'or', 'ta', 'te', 'kn', 'gu', 'hi', 'ml'}
|
21 |
+
|
22 |
+
def detect_language(text):
|
23 |
+
text = text.replace("\n", " ")
|
24 |
+
try:
|
25 |
+
result = detect(text, low_memory=False)
|
26 |
+
detected_lang = result['lang']
|
27 |
+
if detected_lang in SUPPORTED_LANGUAGES:
|
28 |
+
return detected_lang
|
29 |
+
except:
|
30 |
+
pass
|
31 |
+
|
32 |
+
if any('\u0980' <= char <= '\u09FF' for char in text):
|
33 |
+
return 'brx'
|
34 |
+
elif any('\uABC0' <= char <= '\uABFF' for char in text):
|
35 |
+
return 'mni'
|
36 |
+
|
37 |
+
return 'en'
|
38 |
+
|
39 |
+
@router.post("/tts")
|
40 |
+
async def text_to_speech(request: TTSRequest):
|
41 |
+
try:
|
42 |
+
detected_language = detect_language(request.text)
|
43 |
+
|
44 |
+
tts_result = client.tts(
|
45 |
+
request.text,
|
46 |
+
source_language=detected_language,
|
47 |
+
gender=request.gender
|
48 |
+
)
|
49 |
+
|
50 |
+
audio_base64 = tts_result['pipelineResponse'][0]['audio'][0]['audioContent']
|
51 |
+
audio_data = base64.b64decode(audio_base64)
|
52 |
+
|
53 |
+
return StreamingResponse(io.BytesIO(audio_data), media_type="audio/wav")
|
54 |
+
except Exception as e:
|
55 |
+
raise HTTPException(status_code=500, detail=str(e))
|
56 |
+
|
57 |
+
@router.post("/asr")
|
58 |
+
async def speech_to_text(file: UploadFile = File(...), source_language: str = "ml"):
|
59 |
+
try:
|
60 |
+
audio_content = await file.read()
|
61 |
+
asr_result = client.asr(audio_content, source_language=source_language)
|
62 |
+
|
63 |
+
return {"transcription": asr_result}
|
64 |
+
except Exception as e:
|
65 |
+
raise HTTPException(status_code=500, detail=str(e))
|