169 lines
6.1 KiB
Python
169 lines
6.1 KiB
Python
"""Endpointy pro přepis audia.
|
|
|
|
- POST /transcribe/dual → paralelní přepis Deepgram + OpenAI Whisper (dva texty)
|
|
- POST /transcribe/combined → dual + sloučení do jednoho co nejlepšího přepisu (merge)
|
|
|
|
Audio přichází jako proměnná (upload `file` ve form-data). Všechny ostatní parametry
|
|
jsou v POST těle (form fields). API klíče jsou v X- hlavičkách (viz credentials.py).
|
|
Výstup je vždy JSON.
|
|
"""
|
|
import asyncio
|
|
|
|
from fastapi import APIRouter, Depends, File, Form, UploadFile
|
|
from pydantic import BaseModel, Field
|
|
|
|
from ..clients import deepgram_client, openai_client
|
|
from ..config import (
|
|
DEFAULT_CHAT_MODEL,
|
|
DEFAULT_DEEPGRAM_MODEL,
|
|
DEFAULT_LANGUAGE,
|
|
DEFAULT_WHISPER_MODEL,
|
|
)
|
|
from ..credentials import Credentials, get_credentials
|
|
from ..errors import BadRequestError
|
|
from ..logging_config import get_logger
|
|
from ..prompts import DEFAULT_COMBINE_PROMPT
|
|
|
|
log = get_logger("audio-transcription.transcribe")
|
|
|
|
router = APIRouter(tags=["transcribe"])
|
|
|
|
|
|
class DualResult(BaseModel):
|
|
text1: str = Field(..., description="Přepis z Deepgramu.")
|
|
text2: str = Field(..., description="Přepis z OpenAI Whisper.")
|
|
deepgram_model: str
|
|
whisper_model: str
|
|
language: str
|
|
|
|
|
|
class CombinedResult(DualResult):
|
|
merged: str = Field(..., description="Sloučený, co nejpřesnější výsledný přepis.")
|
|
chat_model: str
|
|
|
|
|
|
async def _read_upload(file: UploadFile) -> tuple[bytes, str, str]:
|
|
"""Načte upload do paměti a vrátí (bytes, filename, content_type)."""
|
|
audio = await file.read()
|
|
if not audio:
|
|
raise BadRequestError("Nahraný soubor je prázdný.")
|
|
filename = file.filename or "audio.mp3"
|
|
content_type = file.content_type or "application/octet-stream"
|
|
return audio, filename, content_type
|
|
|
|
|
|
async def _transcribe_dual(
|
|
creds: Credentials,
|
|
audio: bytes,
|
|
filename: str,
|
|
content_type: str,
|
|
deepgram_model: str,
|
|
whisper_model: str,
|
|
language: str,
|
|
diarize: bool,
|
|
smart_format: bool,
|
|
) -> tuple[str, str]:
|
|
"""Spustí oba přepisy paralelně; vrátí (deepgram_text, whisper_text)."""
|
|
deepgram_key = creds.require_deepgram()
|
|
openai_key = creds.require_openai()
|
|
|
|
text1, text2 = await asyncio.gather(
|
|
deepgram_client.transcribe(
|
|
api_key=deepgram_key,
|
|
audio=audio,
|
|
content_type=content_type,
|
|
model=deepgram_model,
|
|
language=language,
|
|
diarize=diarize,
|
|
smart_format=smart_format,
|
|
),
|
|
openai_client.transcribe(
|
|
api_key=openai_key,
|
|
audio=audio,
|
|
filename=filename,
|
|
content_type=content_type,
|
|
model=whisper_model,
|
|
language=language,
|
|
),
|
|
)
|
|
return text1, text2
|
|
|
|
|
|
@router.post(
|
|
"/transcribe/dual",
|
|
response_model=DualResult,
|
|
summary="Paralelní přepis Deepgram + OpenAI Whisper",
|
|
description="Přijme audio soubor a vrátí dva nezávislé přepisy: "
|
|
"`text1` z Deepgramu a `text2` z OpenAI Whisper.",
|
|
)
|
|
async def transcribe_dual(
|
|
file: UploadFile = File(..., description="Audio soubor k přepisu (proměnná)."),
|
|
deepgram_model: str = Form(DEFAULT_DEEPGRAM_MODEL, description="Deepgram model."),
|
|
whisper_model: str = Form(DEFAULT_WHISPER_MODEL, description="OpenAI přepisový model."),
|
|
language: str = Form(DEFAULT_LANGUAGE, description="Jazyk audia (ISO kód, např. cs)."),
|
|
diarize: bool = Form(True, description="Deepgram diarizace (rozlišení mluvčích)."),
|
|
smart_format: bool = Form(True, description="Deepgram smart formatting."),
|
|
creds: Credentials = Depends(get_credentials),
|
|
) -> DualResult:
|
|
audio, filename, content_type = await _read_upload(file)
|
|
text1, text2 = await _transcribe_dual(
|
|
creds, audio, filename, content_type,
|
|
deepgram_model, whisper_model, language, diarize, smart_format,
|
|
)
|
|
return DualResult(
|
|
text1=text1,
|
|
text2=text2,
|
|
deepgram_model=deepgram_model,
|
|
whisper_model=whisper_model,
|
|
language=language,
|
|
)
|
|
|
|
|
|
@router.post(
|
|
"/transcribe/combined",
|
|
response_model=CombinedResult,
|
|
summary="Dual přepis + AI sloučení do nejlepšího přepisu",
|
|
description="Přijme audio soubor, vytvoří přepis z Deepgramu i Whisperu a poté je "
|
|
"sloučí přes OpenAI Chat podle `combine_prompt` do jediného, co nejpřesnějšího "
|
|
"českého přepisu. Vrací `text1`, `text2` i výsledný `merged`.",
|
|
)
|
|
async def transcribe_combined(
|
|
file: UploadFile = File(..., description="Audio soubor k přepisu (proměnná)."),
|
|
combine_prompt: str = Form(
|
|
DEFAULT_COMBINE_PROMPT,
|
|
description="System prompt pro sloučení. Nezadáš-li, použije se výchozí "
|
|
"'Czech Transcript Merger'.",
|
|
),
|
|
deepgram_model: str = Form(DEFAULT_DEEPGRAM_MODEL, description="Deepgram model."),
|
|
whisper_model: str = Form(DEFAULT_WHISPER_MODEL, description="OpenAI přepisový model."),
|
|
chat_model: str = Form(DEFAULT_CHAT_MODEL, description="OpenAI chat model pro sloučení."),
|
|
language: str = Form(DEFAULT_LANGUAGE, description="Jazyk audia (ISO kód, např. cs)."),
|
|
diarize: bool = Form(True, description="Deepgram diarizace (rozlišení mluvčích)."),
|
|
smart_format: bool = Form(True, description="Deepgram smart formatting."),
|
|
creds: Credentials = Depends(get_credentials),
|
|
) -> CombinedResult:
|
|
audio, filename, content_type = await _read_upload(file)
|
|
text1, text2 = await _transcribe_dual(
|
|
creds, audio, filename, content_type,
|
|
deepgram_model, whisper_model, language, diarize, smart_format,
|
|
)
|
|
|
|
prompt = combine_prompt.strip() if combine_prompt and combine_prompt.strip() else DEFAULT_COMBINE_PROMPT
|
|
merged = await openai_client.merge_transcripts(
|
|
api_key=creds.require_openai(),
|
|
model=chat_model,
|
|
system_prompt=prompt,
|
|
text_deepgram=text1,
|
|
text_whisper=text2,
|
|
)
|
|
|
|
return CombinedResult(
|
|
text1=text1,
|
|
text2=text2,
|
|
merged=merged,
|
|
deepgram_model=deepgram_model,
|
|
whisper_model=whisper_model,
|
|
chat_model=chat_model,
|
|
language=language,
|
|
)
|