# Voice

> Add speech transcription to an Angular chat through Copilot Runtime.

`CopilotChat` includes a microphone control. Voice input is transcribed through
Copilot Runtime and inserted into the composer as text.

## Accept voice input [#accept-voice-input]

No component option is required to display the microphone. The browser asks
for microphone permission, records the audio, and sends it to the Runtime
transcription endpoint. The resulting text remains editable before the user
sends it.

Serve production applications over HTTPS so browsers can grant microphone
access.

### Configure Runtime transcription [#configure-runtime-transcription]

Follow the <a href="/angular/voice#backend">Voice backend setup</a> to add a dedicated Runtime
route and transcription service. Keep speech-provider credentials separate
from credentials used for agent chat. For the Built-in Agent showcase, set:

```bash
OPENAI_TRANSCRIPTION_API_KEY=your-api-key
# Optional: defaults to https://api.openai.com/v1
OPENAI_TRANSCRIPTION_BASE_URL=https://your-speech-provider.example.com/v1
```

The route must omit `transcriptionService` when the dedicated credential is
absent. Copilot Runtime then reports `audioFileTranscriptionEnabled: false` on
`/info`, and the chat hides the mic instead of exposing a control that cannot
work. When the service is configured, the browser sends recorded audio to
`{runtimeUrl}/transcribe`. Point `runtimeUrl` in `provideCopilotKit` at the
runtime route that has the transcription service.

Voice requests can call the same backend tools as typed requests. Register
renderers only for the exact tool names your backend exposes. Call
`registerRenderToolCall` from an Angular injection context:

```ts
registerRenderToolCall({
  name: "getWeather",
  args: z.object({ city: z.string() }),
  component: WeatherCardComponent,
});
```

`WeatherCardComponent` is the renderer from
[Render a tool result](/angular/generative-ui/tool-rendering#render-a-tool-result).

The renderer registration is optional. It changes how matching tool calls are
displayed, not how speech is captured or transcribed.

Open the chat. When the runtime has a transcription service, the mic button
shows in the composer. Record a sentence, and the text appears in the input for
you to edit and send.

## Next steps [#next-steps]

- [Multimodal Attachments](/angular/multimodal-attachments)
- [Chat UI configuration](/angular/prebuilt-components)
