Deepgram Voice Workflow

Overview

Use this skill for a complete speech workflow:

1. 1. transcribe audio to text with Deepgram STT
2. optionally synthesize a spoken reply with Deepgram TTS
3. return structured outputs that can feed chat or agent pipelines

This skill is the right choice when the task is broader than plain transcription and needs an input-audio to output-audio pipeline.

Quick Start

Transcribe only

CODEBLOCK0

Generate speech from text

CODEBLOCK1

Run the full pipeline

CODEBLOCK2

Environment

Set DEEPGRAM_API_KEY before use.

The bundled scripts also fall back to reading it from:

• - INLINECODE1

Workflow Decision

Use deepgram-transcribe.sh when

• - only text transcription is needed
• the downstream system will generate its own reply
• the task is speech-to-text only

Use deepgram-tts.sh when

• - text already exists
• only an MP3 spoken response is needed
• the workflow is text-to-speech only

Use neko-voice-pipeline.sh when

• - the task begins with an audio file
• a transcript is needed
• an optional spoken reply should be generated in the same flow

Outputs

STT output

INLINECODE5 writes:

• - transcript text file
• raw API JSON file next to it

TTS output

INLINECODE6 writes:

• - MP3 output file

Pipeline output

INLINECODE7 prints JSON with:

• - INLINECODE8
• INLINECODE9
• INLINECODE10
• INLINECODE11

This makes it easy to wire into scripts or adapters.

Typical Uses

Prefer this skill for:

• - transcribing Telegram/QQ/OneBot voice messages
• generating MP3 replies to short voice prompts
• building bot-side voice input/output automation
• testing speech pipelines from shell without introducing a full SDK

Notes

• - Defaults are tuned for lightweight practical use, not maximal configurability.
• INLINECODE12 defaults to model=nova-2 and language=zh.
• INLINECODE15 defaults to model=aura-2-luna-en; override the model when a different voice is preferred.
• Inspect the raw JSON transcript response when debugging recognition quality or API errors.

References

Read these files when needed:

• - references/stt-notes.md for transcription details
• INLINECODE18 for speech synthesis details
• INLINECODE19 for end-to-end pipeline behavior