How It Works
User sends voice clip in WebEx space
|
v
OpenClaw transcribes automatically (built-in)
|
v
NetClaw processes with full skill set
(pyATS, NetBox, ServiceNow, all 43 MCP servers)
|
v
python3 $MCP_CALL "python3 -u $TTS_MCP_SCRIPT" text_to_speech -> MP3 file
|
v
Upload MP3 to WebEx thread + post text response
Voice Response Workflow
Step 1: Process the question
Treat the transcribed voice message identically to a typed text message. Use the full NetClaw skill set -- pyATS, NetBox, ServiceNow, etc.
Step 2: Generate voice response
After composing your text response, call text_to_speech:
python3 $MCP_CALL "python3 -u $TTS_MCP_SCRIPT" text_to_speech '{"text":"R1 has 3 OSPF neighbors, all in FULL state on Area 0...","voice":"en-US-GuyNeural"}'
This returns JSON with an output_path to the generated MP3 file.
To list available voices:
python3 $MCP_CALL "python3 -u $TTS_MCP_SCRIPT" list_voices '{"language":"en"}'
Step 3: Deliver both text and voice
Post the text response in the WebEx space/thread AND upload the MP3 file as an attachment via the Messages API (multipart/form-data with files parameter):
Voice Response [MP3 audio file attached]
R1 has 3 OSPF neighbors, all in FULL state on Area 0:
- 2.2.2.2 (R2) via Gi1 -- FULL/DR
- 3.3.3.3 (R3) via Gi2 -- FULL/BDR
Always deliver text AND voice. Text is primary (searchable, accessible). Voice is supplementary.
Voice Selection
| Voice | Description |
|---|---|
| en-US-GuyNeural | Professional male -- default |
| en-US-JennyNeural | Professional female |
| en-US-AriaNeural | Conversational female |
| en-GB-RyanNeural | British male |
Users can request a voice change:
- "Switch to a female voice" -> use en-US-JennyNeural
- "Use a British accent" -> use en-GB-RyanNeural
Call list_voices to see all 300+ available voices.
Performance
| Phase | Latency |
|---|---|
| edge-tts synthesis | 1-2 seconds |
| WebEx MP3 upload | < 1 second |
Voice synthesis adds minimal overhead to the response time.
Fallback
If TTS fails, deliver the text response immediately. Do not block on voice.
Tips for Voice Responses
- Keep it concise -- under 100 words works best for spoken delivery
- Avoid tables -- describe data conversationally for voice
- Spell out abbreviations -- say "OSPF" not "O-S-P-F" (edge-tts handles this)
- Use natural phrasing -- the text will be read aloud, so write for the ear
WebEx File Upload for Voice
WebEx Messages API supports file attachments via multipart upload:
POST https://webexapis.com/v1/messages
Content-Type: multipart/form-data
- roomId: <space-id>
- parentId: <thread-parent-message-id> (if threading)
- text: "Voice Response: R1 has 3 OSPF neighbors..."
- files: @/tmp/netclaw-tts/response.mp3
Files up to 100 MB are supported. MP3 voice responses are typically under 1 MB.
GAIT Integration
Record voice interactions in the GAIT audit trail:
Input: Voice clip from @user (transcript: "What are your interfaces?")
Action: Queried R1 interfaces via pyATS
Output: 4 interfaces found -- text + voice response delivered to WebEx