This package streamlines all of the required state management for building client side applications using the EVI Chat WebSocket through a <VoiceProvider> component and useVoice() hook. It provides a WebSocket, Microphone Interface, Audio Playback Queue, and Message History that are all designed to work closely together.
[!NOTE] This package uses Web APIs for microphone input and audio playback that are not compatible with React Native.
[!IMPORTANT] This package is built for use within modern web based React applications using a bundler like
Next.js,Webpack, orVite
Before installing this package, please ensure your development environment meets the following requirement:
- Node.js (
v18.0.0or higher).
To verify your Node.js version, run this command in your terminal:
node --versionIf your Node.js version is below 18.0.0, update it to meet the requirement. For updating Node.js, visit Node.js' official site or use a version management tool like nvm for a more seamless upgrade process.
Add @humeai/voice-react to your project by running this command in your project directory:
npm install @humeai/voice-reactThis will download and include the package in your project, making it ready for import and use within your React components.
import { VoiceProvider } from '@humeai/voice-react';To use the SDK, wrap your components in the VoiceProvider, which will enable your components to access available voice methods. Here's a simple example to get you started:
import { VoiceProvider } from '@humeai/voice-react';
function App() {
const apiKey = process.env.HUME_API_KEY;
return (
<VoiceProvider
auth={{ type: 'apiKey', value: apiKey }}
configId={/* Optional: Your EVI Configuration ID */}
>
{/* ... */}
</VoiceProvider>
);
}Configuring VoiceProvider
See a complete list of props accepted by VoiceProvider below:
(Required) Authentication strategy and corresponding value. Authentication is required to establish the web socket connection with Hume's Voice API. See our documentation on obtaining your API key or access token.
(Optional) Hostname of the Hume API. If not provided this value will default to "api.hume.ai".
(Optional) Number of times to attempt to reconnect to the API. If not provided this value will default to 30.
(Optional) Enable debug mode. If not provided this value will default to false.
(Optional) If you have a configuration ID with voice presets, pass the config ID here.
(Optional) If you wish to use a specific version of your config, pass in the version ID here.
(Optional) A flag to enable verbose transcription. When true, unfinalized user transcripts are sent to the client as interim UserMessage messages, which makes the assistant more sensitive to interruptions. Defaults to true.
onMessage?: (message: JsonMessage & { receivedAt: Date;}) => void
(Optional) Callback function to invoke upon receiving a message through the web socket.
onToolCall?: ToolCallHandler
(Optional) Callback function to invoke upon receiving a ToolCallMessage through the web socket. It will send the string returned as a the content of a ToolResponseMessage. This is where you should add logic that handles your custom tool calls.
(Optional) Callback function to invoke when an audio output message is received from the websocket.
(Optional) Callback function to invoke when an audio clip from the assistant starts playing.
(Optional) Callback function to invoke when an audio clip from the assistant stops playing.
(Optional) Callback function to invoke when the assistant is interrupted.
onClose?: (event: CloseEvent) => void
(Optional) Callback function to invoke upon the web socket connection being closed.
(Optional) Boolean which indicates whether you want to clear message history when the call ends.
(Optional) Set the number of messages that you wish to keep over the course of the conversation. The default value is 100.
sessionSettings?: SessionSettings
(Optional) Settings where you can set custom values for the session
(Optional) Include a chat group ID, which enables the chat to continue from a previous chat group.
After you have set up your voice provider, you will be able to access various properties and methods to use the voice in your application. In any component that is a child of VoiceProvider, access these methods by importing the useVoice custom hook.
For example, to include a button to start a call, you could create a button like this:
import { useVoice } from '@humeai/voice-react';
export function StartCallButton() {
const { connect } = useVoice();
return <button onClick={() => connect()}>Start Call</button>;
}[!IMPORTANT] Under the hood, the React SDK uses the AudioContext API, which must be initialized by a user gesture.
✅ CORRECT: call
connecton a button click.❌ INCORRECT: call
connectin auseEffectto start a call on component mount.
Opens a socket connection to the voice API and initializes the microphone.
| Parameter | Type | Description |
|---|---|---|
options |
ConnectOptions |
Optional settings for the connection. |
Disconnect from the voice API and microphone.
Clear transcript messages from history.
Mute the microphone
Unmute the microphone
Mute the assistant audio
Unmute the assistant audio
Sets the playback volume for audio generated by the assistant. Input values are clamped between 0.0 (silent) and 1.0 (full volume).
sendSessionSettings: (message: SessionSettings) => void
Send new session settings to the assistant. This overrides any session settings that were passed as props to the VoiceProvider.
Send a user input message.
Send a text string for the assistant to read out loud.
sendToolMessage: (toolMessage: ToolResponse | ToolError) => void
Send a tool response or tool error message to the EVI backend.
Pauses responses from EVI. Chat history is still saved and sent after resuming.
Resumes responses from EVI. Chat history sent while paused will now be sent.
Boolean that describes whether the microphone is muted.
Boolean that describes whether the assistant audio is muted.
The current playback volume level for the assistant's voice, ranging from 0.0 (silent) to 1.0 (full volume). Defaults to 1.0.
Describes whether the assistant audio is currently playing.
Boolean that describes whether the assistant is paused. When paused, the assistant will still be listening, but will not send a response until it is resumed.
Audio FFT values for the assistant audio output.
Audio FFT values for microphone input.
messages: UserTranscriptMessage | AssistantTranscriptMessage | ConnectionMessage | UserInterruptionMessage | JSONErrorMessage
Message history of the current conversation. By default, messages does not include interim user messages when verboseTranscription is set to true on the VoiceProvider (verboseTranscription is true by default). To access interim messages, you can define a custom onMessage callback on your VoiceProvider.
lastVoiceMessage: AssistantTranscriptMessage | null
The last transcript message received from the assistant.
lastUserMessage: UserTranscriptMessage | null
The last transcript message received from the user.
readyState: VoiceReadyState
The current readyState of the websocket connection.
status: VoiceStatus
The current status of the voice connection. Informs you of whether the voice is connected, disconnected, connecting, or error. If the voice is in an error state, it will automatically disconnect from the websocket and microphone.
error: VoiceError
Provides more detailed error information if the voice is in an error state.
If true, the voice is in an error state.
If true, an audio playback error has occurred.
If true, a microphone error has occurred.
If true, there was an error connecting to the websocket.
The length of a call. This value persists after the conversation has ended.
toolStatusStore: Record<string, { call?: ToolCall; resolved?: ToolResponse | ToolError }>
A map of tool call IDs to their associated tool messages.
chatMetadata: ChatMetadataMessage | null
Metadata about the current chat, including chat ID, chat group ID, and request ID.
The number of assistant audio clips that are queued up, including the clip that is currently playing.
export type ConnectOptions = {
/** Custom audio constraints passed to navigator.getUserMedia to get the microphone stream */
audioConstraints?: AudioConstraints;
};export type AudioConstraints = {
/** Reduce echo from the input (if supported). Defaults to `true`. */
echoCancellation?: boolean;
/** Suppress background noise (if supported). Defaults to `true`.*/
noiseSuppression?: boolean;
/** Automatically adjust microphone gain (if supported). Defaults to `true`. */
autoGainControl?: boolean;
};If you have questions or require assistance pertaining to this package, reach out to us on Discord!