feat: add audio speech generation endpoint and media handling refactor

- Added new `/v1/audio/speech` and `/api/{path_provider}/audio/speech` endpoints in `g4f/api/__init__.py` for generating speech from text
- Introduced `AudioSpeechConfig` model in `g4f/api/stubs.py` with fields for input, model, provider, voice, instructions, and response format
- Updated `PollinationsAI.py` to support `modalities` in `kwargs` when checking for audio
- Set default voice for audio models in `PollinationsAI.py` if not provided in `kwargs`
- Added debug print in `PollinationsAI.py` to log request data to text API endpoint
- Extended supported FastAPI response types in `g4f/api/__init__.py` to include `FileResponse` from `starlette.responses`
- Added `BackgroundTask` to clean up generated audio files after serving in `g4f/api/__init__.py`
- Modified `AnyProvider.py` to include `EdgeTTS`, `gTTS`, and `MarkItDown` as audio providers when `audio` is in `kwargs` or `modalities`
- Created `resolve_media` helper in `g4f/client/__init__.py` to standardize media handling for audio/image input
- Replaced manual media preprocessing in `Completions`, `AsyncCompletions`, and `Images` classes with `resolve_media`
- Added `/docs/README.md` with a link to the documentation site

docs/README.md

@@ -0,0 +1 @@
+[Documentation](https://gpt4free.github.io/docs/main.html)

g4f/Provider/PollinationsAI.py

@@ -177,7 +177,7 @@ class PollinationsAI(AsyncGeneratorProvider, ProviderModelMixin):
         # Load model list
         cls.get_models()
         if not model:
-            has_audio = "audio" in kwargs
+            has_audio = "audio" in kwargs or "audio" in kwargs.get("modalities", [])
             if not has_audio and media is not None:
                 for media_data, filename in media:
                     if is_data_an_audio(media_data, filename):
@@ -311,6 +311,8 @@ class PollinationsAI(AsyncGeneratorProvider, ProviderModelMixin):
 
         async with ClientSession(headers=DEFAULT_HEADERS, connector=get_connector(proxy=proxy)) as session:
             if model in cls.audio_models:
+                if "audio" in kwargs and kwargs.get("audio", {}).get("voice") is None:
+                    kwargs["audio"]["voice"] = cls.audio_models[model][0]
                 url = cls.text_api_endpoint
                 stream = False
             else:
@@ -329,6 +331,7 @@ class PollinationsAI(AsyncGeneratorProvider, ProviderModelMixin):
                 "cache": cache,
                 **extra_parameters
             })
+            print(f"Requesting {url} with data: {data}")
             async with session.post(url, json=data) as response:
                 await raise_for_status(response)
                 if response.headers["content-type"].startswith("text/plain"):

g4f/api/__init__.py

@@ -13,7 +13,7 @@ import asyncio
 from urllib.parse import quote_plus
 from fastapi import FastAPI, Response, Request, UploadFile, Form, Depends
 from fastapi.middleware.wsgi import WSGIMiddleware
-from fastapi.responses import StreamingResponse, RedirectResponse, HTMLResponse, JSONResponse
+from fastapi.responses import StreamingResponse, RedirectResponse, HTMLResponse, JSONResponse, FileResponse
 from fastapi.exceptions import RequestValidationError
 from fastapi.security import APIKeyHeader
 from starlette.exceptions import HTTPException
@@ -30,6 +30,7 @@ from fastapi.encoders import jsonable_encoder
 from fastapi.security import HTTPBearer, HTTPAuthorizationCredentials, HTTPBasic
 from fastapi.middleware.cors import CORSMiddleware
 from starlette.responses import FileResponse
+from starlette.background import BackgroundTask
 from types import SimpleNamespace
 from typing import Union, Optional, List
 
@@ -49,6 +50,7 @@ from g4f.image.copy_images import get_media_dir, copy_media, get_source_url
 from g4f.errors import ProviderNotFoundError, ModelNotFoundError, MissingAuthError, NoValidHarFileError
 from g4f.cookies import read_cookie_files, get_cookies_dir
 from g4f.providers.types import ProviderType
+from g4f.providers.response import AudioResponse
 from g4f.providers.any_provider import AnyProvider
 from g4f import Provider
 from g4f.gui import get_gui_app
@@ -58,7 +60,7 @@ from .stubs import (
     ProviderResponseModel, ModelResponseModel,
     ErrorResponseModel, ProviderResponseDetailModel,
     FileResponseModel, UploadResponseModel,
-    TranscriptionResponseModel
+    TranscriptionResponseModel, AudioSpeechConfig
 )
 from g4f import debug
 
@@ -492,10 +494,11 @@ class Api:
         }
         @self.app.post("/v1/audio/transcriptions", responses=responses)
         @self.app.post("/api/{path_provider}/audio/transcriptions", responses=responses)
-        async def generate_image(
+        @self.app.post("/api/markitdown", responses=responses)
+        async def convert(
             file: UploadFile,
             model: Annotated[Optional[str], Form()] = None,
-            provider: Annotated[Optional[str], Form()] = AppConfig.media_provider,
+            provider: Annotated[Optional[str], Form()] = "MarkItDown",
             path_provider: str = None,
             prompt: Annotated[Optional[str], Form()] = "Transcribe this audio",
             api_key: Annotated[Optional[str], Form()] = None,
@@ -525,6 +528,54 @@ class Api:
                 logger.exception(e)
                 return ErrorResponse.from_exception(e, None, HTTP_500_INTERNAL_SERVER_ERROR)
 
+        responses = {
+            HTTP_200_OK: {"class": FileResponse},
+            HTTP_401_UNAUTHORIZED: {"model": ErrorResponseModel},
+            HTTP_404_NOT_FOUND: {"model": ErrorResponseModel},
+            HTTP_500_INTERNAL_SERVER_ERROR: {"model": ErrorResponseModel},
+        }
+        @self.app.post("/v1/audio/speech", responses=responses)
+        @self.app.post("/api/{path_provider}/audio/speech", responses=responses)
+        async def generate_speech(
+            config: AudioSpeechConfig,
+            provider: str = AppConfig.media_provider,
+            credentials: Annotated[HTTPAuthorizationCredentials, Depends(Api.security)] = None
+        ):
+            api_key = None
+            if credentials is not None and credentials.credentials != "secret":
+                api_key = credentials.credentials
+            try:
+                response = await self.client.chat.completions.create(
+                    messages=[
+                        {"role": "user", "content": f"{config.instrcutions} Text: {config.input}"}
+                    ],
+                    model=config.model,
+                    provider=config.provider if provider is None else provider,
+                    prompt=config.input,
+                    audio=filter_none(voice=config.voice, format=config.response_format),
+                    **filter_none(
+                        api_key=api_key,
+                    )
+                )
+                if isinstance(response.choices[0].message.content, AudioResponse):
+                    response = response.choices[0].message.content.data
+                    response = response.replace("/media", get_media_dir())
+                    def delete_file():
+                        try:
+                            os.remove(response)
+                        except Exception as e:
+                            logger.exception(e)
+                    return FileResponse(response, background=BackgroundTask(delete_file))
+            except (ModelNotFoundError, ProviderNotFoundError) as e:
+                logger.exception(e)
+                return ErrorResponse.from_exception(e, None, HTTP_404_NOT_FOUND)
+            except MissingAuthError as e:
+                logger.exception(e)
+                return ErrorResponse.from_exception(e, None, HTTP_401_UNAUTHORIZED)
+            except Exception as e:
+                logger.exception(e)
+                return ErrorResponse.from_exception(e, None, HTTP_500_INTERNAL_SERVER_ERROR)
+
         @self.app.post("/v1/upload_cookies", responses={
             HTTP_200_OK: {"model": List[FileResponseModel]},
         })

g4f/api/stubs.py

@@ -118,4 +118,12 @@ class FileResponseModel(BaseModel):
 class TranscriptionResponseModel(BaseModel):
     text: str
     model: str
-    provider: str
+    provider: str
+
+class AudioSpeechConfig(BaseModel):
+    input: str
+    model: Optional[str] = None
+    provider: Optional[str] = None
+    voice: Optional[str] = None
+    instrcutions: str = "Speech this text in a natural way."
+    response_format: Optional[str] = None

g4f/client/__init__.py

@@ -45,6 +45,17 @@ def add_chunk(content, chunk):
         content = str(content) + str(chunk)
     return content
 
+def resolve_media(kwargs: dict, image = None, image_name: str = None) -> None:
+    if image is not None:
+        kwargs["media"] = [(image, image_name)]
+    elif "images" in kwargs:
+        kwargs["media"] = kwargs.pop("images")
+    if "media" in kwargs and not isinstance(kwargs["media"], list):
+        kwargs["media"] = [kwargs["media"]]
+    for idx, media in enumerate(kwargs.get("media", [])):
+        if not isinstance(media, (list, tuple)):
+            kwargs["media"][idx] = (media, getattr(media, "name", None))
+
 # Synchronous iter_response function
 def iter_response(
     response: Union[Iterator[Union[str, ResponseType]]],
@@ -296,13 +307,7 @@ class Completions:
     ) -> ChatCompletion:
         if isinstance(messages, str):
             messages = [{"role": "user", "content": messages}]
-        if image is not None:
+        resolve_media(kwargs, image, image_name)
-            kwargs["media"] = [(image, image_name)]
-        elif "images" in kwargs:
-            kwargs["media"] = kwargs.pop("images")
-        for idx, media in enumerate(kwargs.get("media", [])):
-            if not isinstance(media, (list, tuple)):
-                kwargs["media"][idx] = (media, getattr(media, "name", None))
         if provider is None:
             provider = self.provider
             if provider is None:
@@ -483,6 +488,7 @@ class Images:
     async def async_create_variation(
         self,
         image: ImageType,
+        image_name: str = None,
         model: Optional[str] = None,
         provider: Optional[ProviderType] = None,
         response_format: Optional[str] = None,
@@ -494,11 +500,7 @@ class Images:
         if proxy is None:
             proxy = self.client.proxy
         prompt = "create a variation of this image"
-        if image is not None:
+        resolve_media(kwargs, image, image_name)
-            kwargs["media"] = image
-        for idx, media in enumerate(kwargs.get("media", [])):
-            if not isinstance(media, (list, tuple)):
-                kwargs["media"][idx] = (media, getattr(media, "name", None))
         error = None
         response = None
         if isinstance(provider_handler, IterListProvider):
@@ -600,13 +602,7 @@ class AsyncCompletions:
     ) -> Awaitable[ChatCompletion]:
         if isinstance(messages, str):
             messages = [{"role": "user", "content": messages}]
-        if image is not None:
+        resolve_media(kwargs, image, image_name)
-            kwargs["media"] = [(image, image_name)]
-        elif "images" in kwargs:
-            kwargs["media"] = kwargs.pop("images")
-        for idx, media in enumerate(kwargs.get("media", [])):
-            if not isinstance(media, (list, tuple)):
-                kwargs["media"][idx] = (media, getattr(media, "name", None))
         if provider is None:
             provider = self.provider
             if provider is None:

g4f/providers/any_provider.py

@@ -11,7 +11,7 @@ from ..Provider.hf_space import HuggingSpace
 from .. import Provider
 from .. import models
 from ..Provider import Cloudflare, LMArenaProvider, Gemini, Grok, DeepSeekAPI, PerplexityLabs, LambdaChat, PollinationsAI, FreeRouter
-from ..Provider import Microsoft_Phi_4, DeepInfraChat, Blackbox
+from ..Provider import Microsoft_Phi_4, DeepInfraChat, Blackbox, EdgeTTS, gTTS, MarkItDown
 from .base_provider import AsyncGeneratorProvider, ProviderModelMixin
 
 class AnyProvider(AsyncGeneratorProvider, ProviderModelMixin):
@@ -124,15 +124,17 @@ class AnyProvider(AsyncGeneratorProvider, ProviderModelMixin):
         elif not model or model == cls.default_model:
             model = ""
             has_image = False
-            has_audio = "audio" in kwargs
+            has_audio = False
             if not has_audio and media is not None:
                 for media_data, filename in media:
                     if is_data_an_audio(media_data, filename):
                         has_audio = True
                         break
                     has_image = True
-            if has_audio:
+            if "audio" in kwargs or "audio" in kwargs.get("modalities", []):
-                providers = [PollinationsAI, Microsoft_Phi_4]
+                providers = [PollinationsAI, EdgeTTS, gTTS]
+            elif has_audio:
+                providers = [PollinationsAI, Microsoft_Phi_4, MarkItDown]
             elif has_image:
                 providers = models.default_vision.best_provider.providers
             else: