Skip to content

Gemini live

GeminiLiveVoiceAgent #

Bases: BaseVoiceAgent

Gemini Live Voice Agent.

Source code in llama_index/voice_agents/gemini_live/base.py 
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
class GeminiLiveVoiceAgent(BaseVoiceAgent):
    """
    Gemini Live Voice Agent.
    """

    def __init__(
        self,
        model: Optional[str] = None,
        interface: Optional[GeminiLiveVoiceAgentInterface] = None,
        api_key: Optional[str] = None,
        tools: Optional[List[BaseTool]] = None,
    ):
        self.model: str = model or DEFAULT_MODEL
        self._client: Optional[Client] = None
        self.session: Optional[AsyncSession] = None
        self._quitflag: bool = False
        interface = interface or GeminiLiveVoiceAgentInterface()
        super().__init__(api_key=api_key, tools=tools, interface=interface)
        if self.tools is not None:
            self.gemini_tools: List[Dict[str, List[Dict[str, str]]]] = (
                tools_to_gemini_tools(tools)
            )
            self._functions_dict: Dict[
                str, Callable[[Dict[str, Any], str, str], types.FunctionResponse]
            ] = tools_to_functions_dict(self.tools)
        else:
            self.gemini_tools = []
            self._functions_dict = {}

    @property
    def client(self) -> Client:
        if not self._client:
            self._client = Client(
                api_key=self.api_key, http_options={"api_version": "v1beta"}
            )
        return self._client

    def _signal_exit(self):
        logging.info("Preparing exit...")
        self._quitflag = True

    @override
    async def _start(self, session: AsyncSession) -> None:
        """
        Start the voice agent.
        """
        self.interface.start(session=session)

    async def _run_loop(self) -> None:
        logging.info("The agent is ready for the conversation")
        logging.info("Type q and press enter to stop the conversation at any time")
        while not self._quitflag:
            text = await asyncio.to_thread(
                input,
                "",
            )
            if text == "q":
                self._signal_exit()
            await self.session.send(input=text or ".", end_of_turn=True)
        logging.info("Session has been successfully closed")
        await self.interrupt()
        await self.stop()

    async def send(self) -> None:
        """
        Send audio to the websocket underlying the voice agent.
        """
        while True:
            msg = await self.interface.out_queue.get()
            await self.session.send(input=msg)

    @override
    async def handle_message(self) -> Any:
        """
        Handle incoming message.

        Args:
            message (Any): incoming message (should be dict, but it is kept open also for other types).
            *args: Can take any positional argument.
            **kwargs: Can take any keyword argument.

        Returns:
            out (Any): This function can return any output.

        """
        while True:
            turn = self.session.receive()
            async for response in turn:
                if response.server_content:
                    if data := response.data:
                        await self.interface.receive(data=data)
                        self._messages.append(
                            ChatMessage(
                                role="assistant", blocks=[AudioBlock(audio=data)]
                            )
                        )
                        self._events.append(
                            AudioReceivedEvent(type_t="audio_received", data=data)
                        )
                        continue
                    if text := response.text:
                        self._messages.append(
                            ChatMessage(role="assistant", blocks=[TextBlock(text=text)])
                        )
                        self._events.append(
                            TextReceivedEvent(type_t="text_received", text=text)
                        )
                elif tool_call := response.tool_call:
                    function_responses: List[types.FunctionResponse] = []
                    for fn_call in tool_call.function_calls:
                        self._events.append(
                            ToolCallEvent(
                                type_t="tool_call",
                                tool_name=fn_call.name,
                                tool_args=fn_call.args,
                            )
                        )
                        result = self._functions_dict[fn_call.name](
                            fn_call.args, fn_call.id, fn_call.name
                        )
                        self._events.append(
                            ToolCallResultEvent(
                                type_t="tool_call_result",
                                tool_name=result.name,
                                tool_result=result.response,
                            )
                        )
                        function_responses.append(result)
                    await self.session.send_tool_response(
                        function_responses=function_responses
                    )
            while not self.interface.audio_in_queue.empty():
                await self.interrupt()

    async def start(self):
        try:
            async with (
                self.client.aio.live.connect(
                    model=self.model,
                    config={
                        "response_modalities": ["AUDIO"],
                        "tools": self.gemini_tools,
                    },
                ) as session,
                asyncio.TaskGroup() as tg,
            ):
                self.session = session
                await self._start(session=session)

                _run_loop = tg.create_task(self._run_loop())
                tg.create_task(self.send())
                tg.create_task(self.interface._microphone_callback())
                tg.create_task(self.handle_message())
                tg.create_task(self.interface.output())

                await _run_loop
                raise asyncio.CancelledError("User requested exit")

        except asyncio.CancelledError:
            pass
        except ExceptionGroup as EG:
            await self.stop()

    async def interrupt(self) -> None:
        """
        Interrupt the input/output audio flow.

        Args:
            None
        Returns:
            out (None): This function does not return anything.

        """
        self.interface.interrupt()

    async def stop(self) -> None:
        """
        Stop the conversation with the voice agent.

        Args:
            None
        Returns:
            out (None): This function does not return anything.

        """
        self.interface.stop()

send async #

send() -> None

Send audio to the websocket underlying the voice agent.

Source code in llama_index/voice_agents/gemini_live/base.py 
90
91
92
93
94
95
96
async def send(self) -> None:
    """
    Send audio to the websocket underlying the voice agent.
    """
    while True:
        msg = await self.interface.out_queue.get()
        await self.session.send(input=msg)

handle_message async #

handle_message() -> Any

Handle incoming message.

Parameters:

Name Type Description Default
message Any

incoming message (should be dict, but it is kept open also for other types).

 required
*args

Can take any positional argument.

 required
**kwargs

Can take any keyword argument.

 required

Returns:

Name Type Description
out Any

This function can return any output.
Source code in llama_index/voice_agents/gemini_live/base.py 
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
@override
async def handle_message(self) -> Any:
    """
    Handle incoming message.

    Args:
        message (Any): incoming message (should be dict, but it is kept open also for other types).
        *args: Can take any positional argument.
        **kwargs: Can take any keyword argument.

    Returns:
        out (Any): This function can return any output.

    """
    while True:
        turn = self.session.receive()
        async for response in turn:
            if response.server_content:
                if data := response.data:
                    await self.interface.receive(data=data)
                    self._messages.append(
                        ChatMessage(
                            role="assistant", blocks=[AudioBlock(audio=data)]
                        )
                    )
                    self._events.append(
                        AudioReceivedEvent(type_t="audio_received", data=data)
                    )
                    continue
                if text := response.text:
                    self._messages.append(
                        ChatMessage(role="assistant", blocks=[TextBlock(text=text)])
                    )
                    self._events.append(
                        TextReceivedEvent(type_t="text_received", text=text)
                    )
            elif tool_call := response.tool_call:
                function_responses: List[types.FunctionResponse] = []
                for fn_call in tool_call.function_calls:
                    self._events.append(
                        ToolCallEvent(
                            type_t="tool_call",
                            tool_name=fn_call.name,
                            tool_args=fn_call.args,
                        )
                    )
                    result = self._functions_dict[fn_call.name](
                        fn_call.args, fn_call.id, fn_call.name
                    )
                    self._events.append(
                        ToolCallResultEvent(
                            type_t="tool_call_result",
                            tool_name=result.name,
                            tool_result=result.response,
                        )
                    )
                    function_responses.append(result)
                await self.session.send_tool_response(
                    function_responses=function_responses
                )
        while not self.interface.audio_in_queue.empty():
            await self.interrupt()

interrupt async #

interrupt() -> None

Interrupt the input/output audio flow.

Returns: out (None): This function does not return anything.

Source code in llama_index/voice_agents/gemini_live/base.py 
190
191
192
193
194
195
196
197
198
199
200
async def interrupt(self) -> None:
    """
    Interrupt the input/output audio flow.

    Args:
        None
    Returns:
        out (None): This function does not return anything.

    """
    self.interface.interrupt()

stop async #

stop() -> None

Stop the conversation with the voice agent.

Returns: out (None): This function does not return anything.

Source code in llama_index/voice_agents/gemini_live/base.py 
202
203
204
205
206
207
208
209
210
211
212
async def stop(self) -> None:
    """
    Stop the conversation with the voice agent.

    Args:
        None
    Returns:
        out (None): This function does not return anything.

    """
    self.interface.stop()

GeminiLiveVoiceAgentInterface #

Bases: BaseVoiceAgentInterface

Source code in llama_index/voice_agents/gemini_live/audio_interface.py 
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
class GeminiLiveVoiceAgentInterface(BaseVoiceAgentInterface):
    def __init__(self) -> None:
        self.audio_in_queue: Optional[asyncio.Queue] = None
        self.out_queue: Optional[asyncio.Queue] = None

        self.session: Optional[AsyncSession] = None
        self.audio_stream: Optional[pyaudio.Stream] = None

    def _speaker_callback(self, *args: Any, **kwargs: Any) -> Any:
        """
        Callback function for the audio output device.

        Args:
            *args: Can take any positional argument.
            **kwargs: Can take any keyword argument.

        Returns:
            out (Any): This function can return any output.

        """

    @override
    async def _microphone_callback(self) -> None:
        """
        Callback function for the audio input device.

        Args:
            *args: Can take any positional argument.
            **kwargs: Can take any keyword argument.

        Returns:
            out (Any): This function can return any output.

        """
        mic_info = pya.get_default_input_device_info()
        self.audio_stream = await asyncio.to_thread(
            pya.open,
            format=FORMAT,
            channels=CHANNELS,
            rate=SEND_SAMPLE_RATE,
            input=True,
            input_device_index=mic_info["index"],
            frames_per_buffer=CHUNK_SIZE,
        )
        if __debug__:
            kwargs = {"exception_on_overflow": False}
        else:
            kwargs = {}
        while True:
            data = await asyncio.to_thread(self.audio_stream.read, CHUNK_SIZE, **kwargs)
            await self.out_queue.put({"data": data, "mime_type": "audio/pcm"})

    @override
    def start(self, session: AsyncSession) -> None:
        """
        Start the interface.

        Args:
            session (AsyncSession): the session to which the API is bound.

        """
        self.session = session
        self.audio_in_queue = asyncio.Queue()
        self.out_queue = asyncio.Queue(maxsize=5)

    def stop(self) -> None:
        """
        Stop the interface.

        Args:
            None
        Returns:
            out (None): This function does not return anything.

        """
        if self.audio_stream:
            self.audio_stream.close()
        else:
            raise ValueError("Audio stream has never been opened, cannot be closed.")

    def interrupt(self) -> None:
        """
        Interrupt the interface.

        Args:
            None
        Returns:
            out (None): This function does not return anything.

        """
        self.audio_in_queue.get_nowait()

    @override
    async def output(self, *args: Any, **kwargs: Any) -> Any:
        """
        Process and output the audio.

        Args:
            *args: Can take any positional argument.
            **kwargs: Can take any keyword argument.

        Returns:
            out (Any): This function can return any output.

        """
        stream = await asyncio.to_thread(
            pya.open,
            format=FORMAT,
            channels=CHANNELS,
            rate=RECEIVE_SAMPLE_RATE,
            output=True,
        )
        while True:
            bytestream = await self.audio_in_queue.get()
            await asyncio.to_thread(stream.write, bytestream)

    @override
    async def receive(self, data: bytes) -> Any:
        """
        Receive audio data.

        Args:
            data (Any): received audio data (generally as bytes or str, but it is kept open also to other types).
            *args: Can take any positional argument.
            **kwargs: Can take any keyword argument.

        Returns:
            out (Any): This function can return any output.

        """
        self.audio_in_queue.put_nowait(data)

start #

start(session: AsyncSession) -> None

Start the interface.

Parameters:

Name Type Description Default
session AsyncSession

the session to which the API is bound.

 required
Source code in llama_index/voice_agents/gemini_live/audio_interface.py 
71
72
73
74
75
76
77
78
79
80
81
82
@override
def start(self, session: AsyncSession) -> None:
    """
    Start the interface.

    Args:
        session (AsyncSession): the session to which the API is bound.

    """
    self.session = session
    self.audio_in_queue = asyncio.Queue()
    self.out_queue = asyncio.Queue(maxsize=5)

stop #

stop() -> None

Stop the interface.

Returns: out (None): This function does not return anything.

Source code in llama_index/voice_agents/gemini_live/audio_interface.py 
84
85
86
87
88
89
90
91
92
93
94
95
96
97
def stop(self) -> None:
    """
    Stop the interface.

    Args:
        None
    Returns:
        out (None): This function does not return anything.

    """
    if self.audio_stream:
        self.audio_stream.close()
    else:
        raise ValueError("Audio stream has never been opened, cannot be closed.")

interrupt #

interrupt() -> None

Interrupt the interface.

Returns: out (None): This function does not return anything.

Source code in llama_index/voice_agents/gemini_live/audio_interface.py 
 99
100
101
102
103
104
105
106
107
108
109
def interrupt(self) -> None:
    """
    Interrupt the interface.

    Args:
        None
    Returns:
        out (None): This function does not return anything.

    """
    self.audio_in_queue.get_nowait()

output async #

output(*args: Any, **kwargs: Any) -> Any

Process and output the audio.

Parameters:

Name Type Description Default
*args Any

Can take any positional argument.

 ()
**kwargs Any

Can take any keyword argument.

 {}

Returns:

Name Type Description
out Any

This function can return any output.
Source code in llama_index/voice_agents/gemini_live/audio_interface.py 
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
@override
async def output(self, *args: Any, **kwargs: Any) -> Any:
    """
    Process and output the audio.

    Args:
        *args: Can take any positional argument.
        **kwargs: Can take any keyword argument.

    Returns:
        out (Any): This function can return any output.

    """
    stream = await asyncio.to_thread(
        pya.open,
        format=FORMAT,
        channels=CHANNELS,
        rate=RECEIVE_SAMPLE_RATE,
        output=True,
    )
    while True:
        bytestream = await self.audio_in_queue.get()
        await asyncio.to_thread(stream.write, bytestream)

receive async #

receive(data: bytes) -> Any

Receive audio data.

Parameters:

Name Type Description Default
data Any

received audio data (generally as bytes or str, but it is kept open also to other types).

 required
*args

Can take any positional argument.

 required
**kwargs

Can take any keyword argument.

 required

Returns:

Name Type Description
out Any

This function can return any output.
Source code in llama_index/voice_agents/gemini_live/audio_interface.py 
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
@override
async def receive(self, data: bytes) -> Any:
    """
    Receive audio data.

    Args:
        data (Any): received audio data (generally as bytes or str, but it is kept open also to other types).
        *args: Can take any positional argument.
        **kwargs: Can take any keyword argument.

    Returns:
        out (Any): This function can return any output.

    """
    self.audio_in_queue.put_nowait(data)

options: members: - GeminiLiveVoiceAgent