Description

This node is responsible for handling Kuri voice interactions. Specifically, audio_voice_delegate has two responsibilities:

• determining if a wake word has been spoken
• managing audio transcription and voice command matching

If a wake word is detected, a request is made to the Houndify service to determine the spoken command, before publishing the results on a ROS topic.

Dependencies

• PulseAudio
• soundhound_kuri

Subscribed Topics

• /image_wp_server/waypoint_update
• /voice_commands/update

Published Topics

/audio/voice_delegate/asleep

Asleep.msg
If audio enters the “asleep” state as a result of the awake_timeout (while transcribing audio) or a call to /audio/voice_delegate/sleep, an empty message will be published

/audio/voice_delegate/awake

Awake.msg
If audio enters the “awake” state as a result of a wake word being detected or the service /audio/voice_delegate/wake_up is called, a message will be published. If the wake word triggered this state change, a non-zero direction vector will be provided for the observed source of audio.

$ rostopic echo /audio/voice_delegate/awake
direction:
  x: -97.0
  y: 26.0
  z: 0.0
---

/audio/voice_delegate/exchange

Exchange.msg
If an audio transcription or command is received from Houndify, a message will be published containing the matched command(s), the raw transcription, and any known error messages

$ rostopic echo /audio/voice_delegate/exchange
commands:
  -
    name: CustomCommand
    params:
      -
        k: name
        v: Happy Birthday Song
error: ''
transcription: it's my birthday
---
commands:
  -
    name: StopCommand
    params: []
error: ''
transcription: stop
---
commands:
  -
    name: TurnCommand
    params:
      -
        k: direction
        v: -90
error: ''
transcription: turn left
---

Services

/audio/voice_delegate/deafen

Deafen.srv
Sets mayfield_audio to the “deaf” state. This is the most common way to temporarily set Kuri to ignore wake words and voice commands. Kuri must be returned to the “asleep” state via service call in order for wake words to be recognized.

$ rosservice call /audio/voice_delegate/deafen
state: deaf

/audio/voice_delegate/snooze

Snooze.srv
Sets mayfield_audio to the “asleep” listening state. This effectively ends any ongoing voice command transcriptions and returns Kuri to a state of listening for the wake word.

$ rosservice call /audio/voice_delegate/snooze
state: asleep

/audio/voice_delegate/stat

Stat.srv
Reports the current audio status of mayfield_audio. Possible states include “awake” (transcribing audio), “asleep” (listening for wake words), and “deaf” (ignoring all audio inputs).

$ rosservice call /audio/voice_delegate/stat
state: asleep
direction: -97.000000, 26.000000, 0.000000

/audio/voice_delegate/wake_up

WakeUp.srv
Sets mayfield_audio to the “awake” state. This service triggers Kuri to start transcribing audio in attempt to find a voice command match. Transcription will time out after /audio_voice_delegate/awake_timeout and Kuri will return to the “asleep” state (listening for wake word).

$ rosservice call /audio/voice_delegate/wake_up
state: awake

Service Calls

• /speaker_phone/get_direction

Parameters

/audio_voice_delegate/agent/params/client_id

The client ID provided by Houndify for identification with an account and subscribed domains. Defaults to the Mayfield client/account

/audio_voice_delegate/agent/params/client_key

The client key associated with the Houndify client ID, above. Defaults to the Mayfield client key

/audio_voice_delegate/agent/params/user_id

A unique identifier for the robot, used to authenticate the robot with the Houndify service, and provide improved location name matching for “go to” voice commands. Defaults to the Kuri UUID

/audio_voice_delegate/agent/type

Identifier for the voice transcription service to use. It is recommended that this be set to soundhound and not modified for the stability of voice commands.

/audio_voice_delegate/awake_timeout

The duration of audio samples to send before ending transcription, measured in seconds

/audio_voice_delegate/capture/params/channels

The number of channels associated with the audio buffer used for listening

/audio_voice_delegate/capture/params/device

PulseAudio source or sink-monitor associated with the processed audio input, used for wake word detection and audio transcription

/audio_voice_delegate/capture/params/latency_msecs

The acceptable number of milliseconds of latency, used to configure the stream for listening to audio

/audio_voice_delegate/capture/params/sample_rate

The sample rate of the audio buffer used for listening

/audio_voice_delegate/capture/type

System audio API type. Defaults to pulse

/audio_voice_delegate/debug

Deprecated (unused)

/audio_voice_delegate/direction

Deprecated (unused)

/audio_voice_delegate/wake/type

Identifier for the wake word library to use. It is recommended that this be set to okhound and not modified for the stability of wake word detection.

/audio_voice_delegate/wake_threshold

Deprecated (unused)

Launch File

audio_voice_delegate.launch