feat: Add voice input and output support

This change adds support for voice input and output to the GPTbot. Users can enable this feature using the new `!gptbot roomsettings` command. Voice input and output are currently supported via OpenAI's TTS and Whisper models. However, note that voice input may be unreliable at the moment. This enhancement expands the capabilities of the bot, allowing users to interact with it using their voice. This addresses the need for a more user-friendly and natural way of communication.
This commit is contained in:
Kumi 2023-11-26 07:58:10 +01:00
parent 554d3d8aa0
commit 155ea68e7a
Signed by: kumi
GPG key ID: ECBCC9082395383F
9 changed files with 227 additions and 20 deletions

View file

@ -1,10 +1,10 @@
# GPTbot
GPTbot is a simple bot that uses different APIs to generate responses to
GPTbot is a simple bot that uses different APIs to generate responses to
messages in a Matrix room.
It is called GPTbot because it was originally intended to only use GPT-3 to
generate responses. However, it supports other services/APIs, and I will
generate responses. However, it supports other services/APIs, and I will
probably add more in the future, so the name is a bit misleading.
## Features
@ -12,9 +12,12 @@ probably add more in the future, so the name is a bit misleading.
- AI-generated responses to messages in a Matrix room (chatbot)
- Currently supports OpenAI (tested with `gpt-3.5-turbo` and `gpt-4`)
- AI-generated pictures via the `!gptbot imagine` command
- Currently supports OpenAI (DALL-E)
- Currently supports OpenAI (DALL-E-2/DALL-E-3)
- Mathematical calculations via the `!gptbot calculate` command
- Currently supports WolframAlpha
- Voice input and output
- Currently supports OpenAI (TTS and Whisper)
- Beta feature, see dedicated section for details
- Automatic classification of messages (for `imagine`, `calculate`, etc.)
- Beta feature, see Usage section for details
- Really useful commands like `!gptbot help` and `!gptbot coin`
@ -26,9 +29,9 @@ probably add more in the future, so the name is a bit misleading.
## Installation
To run the bot, you will need Python 3.10 or newer.
To run the bot, you will need Python 3.10 or newer.
The bot has been tested with Python 3.11 on Arch, but should work with any
The bot has been tested with Python 3.11 on Arch, but should work with any
current version, and should not require any special dependencies or operating
system features.
@ -53,7 +56,7 @@ A release to PyPI is planned, but not yet available.
### Development
Clone the repository and install the requirements to a virtual environment.
Clone the repository and install the requirements to a virtual environment.
```shell
# Clone the repository
@ -145,6 +148,14 @@ Also note that this feature conflicts with the `always_reply false` setting -
or rather, it doesn't make sense then because you already have to explicitly
specify the command to use.
## Voice input and output
The bot supports voice input and output, but it is disabled by default. To
enable it, use the `!gptbot roomsettings` command to change the settings for
the current room. `!gptbot roomsettings stt true` will enable voice input,
and `!gptbot roomsettings tts true` will enable voice output. Note that this
may be a little unreliable at the moment, especially voice input.
## Troubleshooting
**Help, the bot is not responding!**
@ -181,4 +192,5 @@ please check the logs and open an issue if you can't figure out what's going on.
## License
This project is licensed under the terms of the MIT license. See the [LICENSE](LICENSE) file for details.
This project is licensed under the terms of the MIT license. See the [LICENSE](LICENSE)
file for details.

View file

@ -7,7 +7,7 @@ allow-direct-references = true
[project]
name = "matrix-gptbot"
version = "0.2.0"
version = "0.2.1"
authors = [
{ name="Kumi Mitterer", email="gptbot@kumi.email" },
@ -39,6 +39,7 @@ dependencies = [
[project.optional-dependencies]
openai = [
"openai>=1.2",
"pydub",
]
wolframalpha = [

View file

@ -6,5 +6,6 @@ duckdb
python-magic
pillow
wolframalpha
pydub
git+https://kumig.it/kumitterer/trackingmore-api-tool.git

View file

@ -83,6 +83,8 @@ class GPTBot:
chat_api: Optional[OpenAI] = None
image_api: Optional[OpenAI] = None
classification_api: Optional[OpenAI] = None
tts_api: Optional[OpenAI] = None
stt_api: Optional[OpenAI] = None
parcel_api: Optional[TrackingMore] = None
operator: Optional[str] = None
room_ignore_list: List[str] = [] # List of rooms to ignore invites from
@ -149,9 +151,14 @@ class GPTBot:
if "AllowedUsers" in config["GPTBot"]:
bot.allowed_users = json.loads(config["GPTBot"]["AllowedUsers"])
bot.chat_api = bot.image_api = bot.classification_api = OpenAI(
bot, config["OpenAI"]["APIKey"], config["OpenAI"].get("Model"),
config["OpenAI"].get("ImageModel"), config["OpenAI"].get("BaseURL"), bot.logger
bot.chat_api = bot.image_api = bot.classification_api = bot.tts_api = bot.stt_api = OpenAI(
bot=bot,
api_key=config["OpenAI"]["APIKey"],
chat_model=config["OpenAI"].get("Model"),
image_model=config["OpenAI"].get("ImageModel"),
tts_model=config["OpenAI"].get("TTSModel"),
stt_model=config["OpenAI"].get("STTModel"),
base_url=config["OpenAI"].get("BaseURL")
)
bot.max_tokens = config["OpenAI"].getint("MaxTokens", bot.max_tokens)
bot.max_messages = config["OpenAI"].getint("MaxMessages", bot.max_messages)
@ -207,7 +214,7 @@ class GPTBot:
return user_id
async def _last_n_messages(self, room: str | MatrixRoom, n: Optional[int], ignore_bot_commands: bool = True):
async def _last_n_messages(self, room: str | MatrixRoom, n: Optional[int], ignore_bot_commands: bool = False):
messages = []
n = n or self.max_messages
room_id = room.room_id if isinstance(room, MatrixRoom) else room
@ -264,8 +271,7 @@ class GPTBot:
messages.append(event)
if isinstance(event, RoomMessageMedia):
if event.sender != self.matrix_client.user_id:
messages.append(event)
messages.append(event)
self.logger.log(f"Found {len(messages)} messages (limit: {n})", "debug")
@ -574,6 +580,39 @@ class GPTBot:
self.logger.log("Sent image", "debug")
async def send_file(
self, room: MatrixRoom, file: bytes, filename: str, mime: str, msgtype: str
):
"""Send a file to a room.
Args:
room (MatrixRoom): The room to send the file to.
file (bytes): The file to send.
filename (str): The name of the file.
mime (str): The MIME type of the file.
"""
self.logger.log(
f"Sending file of size {len(file)} bytes to room {room.room_id}", "debug"
)
content_uri = await self.upload_file(file, filename, mime)
self.logger.log("Uploaded file - sending message...", "debug")
content = {
"body": filename,
"info": {"mimetype": mime, "size": len(file)},
"msgtype": msgtype,
"url": content_uri,
}
status = await self.matrix_client.room_send(
room.room_id, "m.room.message", content
)
self.logger.log("Sent file", "debug")
async def send_message(
self, room: MatrixRoom | str, message: str, notice: bool = False
):
@ -861,6 +900,46 @@ class GPTBot:
space,
)
def room_uses_stt(self, room: MatrixRoom | str) -> bool:
"""Check if a room uses STT.
Args:
room (MatrixRoom | str): The room to check.
Returns:
bool: Whether the room uses STT.
"""
room_id = room.room_id if isinstance(room, MatrixRoom) else room
with closing(self.database.cursor()) as cursor:
cursor.execute(
"SELECT value FROM room_settings WHERE room_id = ? AND setting = ?",
(room_id, "stt"),
)
result = cursor.fetchone()
return False if not result else bool(int(result[0]))
def room_uses_tts(self, room: MatrixRoom | str) -> bool:
"""Check if a room uses TTS.
Args:
room (MatrixRoom | str): The room to check.
Returns:
bool: Whether the room uses TTS.
"""
room_id = room.room_id if isinstance(room, MatrixRoom) else room
with closing(self.database.cursor()) as cursor:
cursor.execute(
"SELECT value FROM room_settings WHERE room_id = ? AND setting = ?",
(room_id, "tts"),
)
result = cursor.fetchone()
return False if not result else bool(int(result[0]))
def respond_to_room_messages(self, room: MatrixRoom | str) -> bool:
"""Check whether the bot should respond to all messages sent in a room.
@ -955,7 +1034,25 @@ class GPTBot:
message_body = message.body if not self.chat_api.supports_chat_images() else [{"type": "text", "text": message.body}]
chat_messages.append({"role": role, "content": message_body})
if self.chat_api.supports_chat_images() and isinstance(message, RoomMessageMedia):
if isinstance(message, RoomMessageAudio):
role = (
"assistant" if message.sender == self.matrix_client.user_id else "user"
)
if message == event or (not message.event_id == event.event_id):
if self.room_uses_stt(room):
try:
download = await self.download_file(message.url)
message_text = await self.stt_api.speech_to_text(download.body)
except Exception as e:
self.logger.log(f"Error generating text from audio: {e}", "error")
message_text = message.body
else:
message_text = message.body
message_body = message_text if not self.chat_api.supports_chat_images() else [{"type": "text", "text": message_text}]
chat_messages.append({"role": role, "content": message_body})
if self.chat_api.supports_chat_images() and isinstance(message, RoomMessageImage):
image_url = message.url
download = await self.download_file(image_url)
@ -1001,6 +1098,20 @@ class GPTBot:
self.logger.log(f"Sending response to room {room.room_id}...")
if self.room_uses_tts(room):
self.logger.log("TTS enabled for room", "debug")
try:
audio = await self.tts_api.text_to_speech(response)
await self.send_file(room, audio, response, "audio/mpeg", "m.audio")
return
except Exception as e:
self.logger.log(f"Error generating audio: {e}", "error")
await self.send_message(
room, "Something went wrong generating audio file.", True
)
message = await self.send_message(room, response)
else:

View file

@ -3,13 +3,16 @@ import requests
import asyncio
import json
from functools import partial
from contextlib import closing
from typing import Dict, List, Tuple, Generator, AsyncGenerator, Optional, Any
from io import BytesIO
from pydub import AudioSegment
from .logging import Logger
from typing import Dict, List, Tuple, Generator, AsyncGenerator, Optional, Any
ASSISTANT_CODE_INTERPRETER = [
{
"type": "code_interpreter",
@ -30,17 +33,23 @@ class OpenAI:
classification_api = chat_api
image_model: str = "dall-e-2"
tts_model: str = "tts-1-hd"
tts_voice: str = "alloy"
stt_model: str = "whisper-1"
operator: str = "OpenAI ([https://openai.com](https://openai.com))"
def __init__(self, bot, api_key, chat_model=None, image_model=None, base_url=None, logger=None):
def __init__(self, bot, api_key, chat_model=None, image_model=None, tts_model=None, tts_voice=None, stt_model=None, base_url=None, logger=None):
self.bot = bot
self.api_key = api_key
self.chat_model = chat_model or self.chat_model
self.image_model = image_model or self.image_model
self.logger = logger or Logger()
self.logger = logger or bot.logger or Logger()
self.base_url = base_url or openai.base_url
self.openai_api = openai.AsyncOpenAI(api_key=self.api_key, base_url=self.base_url)
self.tts_model = tts_model or self.tts_model
self.tts_voice = tts_voice or self.tts_voice
self.stt_model = stt_model or self.stt_model
def supports_chat_images(self):
return "vision" in self.chat_model
@ -266,6 +275,47 @@ Only the event_types mentioned above are allowed, you must not respond in any ot
return result, tokens_used
async def text_to_speech(self, text: str, user: Optional[str] = None) -> Generator[bytes, None, None]:
"""Generate speech from text.
Args:
text (str): The text to use.
Yields:
bytes: The audio data.
"""
self.logger.log(f"Generating speech from text '{text}'...")
speech = await self.openai_api.audio.speech.create(
model=self.tts_model,
input=text,
voice=self.tts_voice
)
return speech.content
async def speech_to_text(self, audio: bytes, user: Optional[str] = None) -> Tuple[str, int]:
"""Generate text from speech.
Args:
audio (bytes): The audio data.
Returns:
Tuple[str, int]: The text and the number of tokens used.
"""
self.logger.log(f"Generating text from speech...")
response = await self.openai_api.audio.transcriptions.create(
model=self.stt_model,
file=BytesIO(audio),
)
text = response.text
self.logger.log(f"Generated text with {tokens_used} tokens.")
return text
async def generate_image(self, prompt: str, user: Optional[str] = None) -> Generator[bytes, None, None]:
"""Generate an image from a prompt.

View file

@ -22,6 +22,7 @@ for command in [
"dice",
"parcel",
"space",
"tts",
]:
function = getattr(import_module(
"." + command, "gptbot.commands"), "command_" + command)

View file

@ -19,6 +19,7 @@ async def command_help(room: MatrixRoom, event: RoomMessageText, bot):
- !gptbot chat \<message\> - Send a message to the chat API
- !gptbot classify \<message\> - Classify a message using the classification API
- !gptbot custom \<message\> - Used for custom commands handled by the chat model and defined through the room's system message
- !gptbot roomsettings [use_classification|use_timing|always_reply|system_message|tts] [true|false|\<message\>] - Get or set room settings
- !gptbot ignoreolder - Ignore messages before this point as context
"""

View file

@ -25,6 +25,8 @@ async def command_roomsettings(room: MatrixRoom, event: RoomMessageText, bot):
(room.room_id, "system_message", value, value)
)
bot.database.commit()
await bot.send_message(room, f"Alright, I've stored the system message: '{value}'.", True)
return
@ -35,7 +37,7 @@ async def command_roomsettings(room: MatrixRoom, event: RoomMessageText, bot):
await bot.send_message(room, f"The current system message is: '{system_message}'.", True)
return
if setting in ("use_classification", "always_reply", "use_timing"):
if setting in ("use_classification", "always_reply", "use_timing", "tts", "stt"):
if value:
if value.lower() in ["true", "false"]:
value = value.lower() == "true"
@ -49,6 +51,8 @@ async def command_roomsettings(room: MatrixRoom, event: RoomMessageText, bot):
(room.room_id, setting, "1" if value else "0", "1" if value else "0")
)
bot.database.commit()
await bot.send_message(room, f"Alright, I've set {setting} to: '{value}'.", True)
return
@ -81,6 +85,9 @@ async def command_roomsettings(room: MatrixRoom, event: RoomMessageText, bot):
- system_message [message]: Get or set the system message to be sent to the chat model
- classification [true/false]: Get or set whether the room uses classification
- always_reply [true/false]: Get or set whether the bot should reply to all messages (if false, only reply to mentions and commands)
- tts [true/false]: Get or set whether the bot should generate audio files instead of sending text
- stt [true/false]: Get or set whether the bot should attempt to process information from audio files
- timing [true/false]: Get or set whether the bot should return information about the time it took to generate a response
"""
await bot.send_message(room, message, True)

View file

@ -0,0 +1,23 @@
from nio.events.room_events import RoomMessageText
from nio.rooms import MatrixRoom
async def command_tts(room: MatrixRoom, event: RoomMessageText, bot):
prompt = " ".join(event.body.split()[2:])
if prompt:
bot.logger.log("Generating speech...")
try:
content = await bot.tts_api.text_to_speech(prompt, user=room.room_id)
except Exception as e:
bot.logger.log(f"Error generating speech: {e}", "error")
await bot.send_message(room, "Sorry, I couldn't generate an audio file. Please try again later.", True)
return
bot.logger.log(f"Sending audio file...")
await bot.send_file(room, content, "audio.mp3", "audio/mpeg", "m.audio")
return
await bot.send_message(room, "You need to provide a prompt.", True)