Powered byLightning AI
All blueprintsEngineering AI workflow blueprint

Speech-to-text and text-to-speech workflow blueprint

Synthesize text to speech, transcribe it back, and measure accuracy -- showcasing end-to-end voice AI

See where this blueprint fits in the production workflow lifecycle
Tts
Asr
Speech
Multilingual
Voice Ai
Use this blueprint
Pattern
Dag
Agents
1
Functions
3
MCP servers
0

What the Speech Roundtrip (TTS + ASR) blueprint does

A pipeline that demonstrates the full speech loop: text-to-speech synthesis with Qwen3-TTS, automatic speech recognition with Qwen3-ASR, LLM-based normalization, and character-level accuracy comparison. Supports Japanese and other languages. **Architecture:** - **Synthesize**: Converts input text to audio using Qwen3-TTS (vllm-omni) - **Transcribe**: Runs ASR on the generated audio using Qwen3-ASR - **Normalize**: An LLM agent cleans up ASR output (e.g. romanized text back to native script) - **Compare**: Character-level accuracy measurement using longest common subsequence **Input:** - `text`: The text to synthesize and round-trip - `voice` (optional): Speaker voice name - `instructions` (optional): Style instructions for TTS Perfect for: - Evaluating TTS + ASR pipeline quality - Multilingual speech processing demos - Voice AI integration testing

Inputs for the Speech Roundtrip (TTS + ASR) workflow

Provide the following information when starting this workflow. Required inputs are validated before the first workflow stage runs.

  • Text

    Required

    Text to synthesize and round-trip

    Type: String

  • Language

    Optional

    TTS language (e.g. Japanese, English, Chinese)

    Type: StringSource default or example: Japanese

  • Voice

    Optional

    Speaker voice name (optional)

    Type: String

  • Instructions

    Optional

    Style instructions for TTS (optional)

    Type: String

Input template

This request shape is generated from the blueprint's input schema. Angle-bracket values are placeholders; defaults and examples appear only when the source blueprint defines them.

{
  "text": "<required string>",
  "language": "Japanese",
  "voice": "<optional string>",
  "instructions": "<optional string>"
}

How the Speech Roundtrip (TTS + ASR) AI workflow works

This engineering blueprint uses a dag workflow with 1 agent, 3 functions, and 0 MCP servers. Open it in GraphN to customize the resources, inputs, and workflow behavior for your use case.

  1. Text to Speech

    Function using Text To Speech

  2. Speech to Text

    Function using Speech To Text

  3. Normalize ASR

    Agent using Normalizer

  4. Compare

    Function using Compare Results

Blueprint resources and prerequisites

These are the actual agents, functions, and MCP servers declared by the blueprint. If an MCP server connects to an external system, configure the required connection and credentials in your GraphN workspace before running the workflow.

AI agent

Normalizer

You are a text normalization assistant.

Function

Text To Speech

Synthesize text to audio using Qwen3-TTS

Function

Speech To Text

Transcribe audio using Qwen3-ASR

Function

Compare Results

Compare original text with transcription using character-level accuracy

Workflow output contract

The completed workflow makes the following output available to downstream steps, applications, or API clients:

Compare

Frequently asked questions

What does the Speech Roundtrip (TTS + ASR) blueprint do?
Synthesize text to speech, transcribe it back, and measure accuracy -- showcasing end-to-end voice AI
What inputs does the Speech Roundtrip (TTS + ASR) workflow use?
The workflow uses Text, Language, Voice, Instructions. Required fields and their expected formats are listed on this page.
What resources are included in the Speech Roundtrip (TTS + ASR) blueprint?
The blueprint contains 1 AI agent, 3 functions, 0 MCP servers. Its workflow pattern is Dag.
Can I customize the Speech Roundtrip (TTS + ASR) blueprint in GraphN?
Yes. Open the blueprint in GraphN to customize its inputs, models, resources, and workflow behavior before deployment.

Related Engineering blueprints

View all software engineering workflows

API Data Summarizer

Fetch any public JSON API and get a clear, plain-language summary of the response

API Docs Generator

Point it at your API code and get OpenAPI-style docs with examples

Architecture Decision Record Writer

Describe a technical decision, get a formatted ADR with tradeoffs and alternatives

Production Incident Response

Triage a production incident: analyze the alert, logs, and evidence in parallel and propose next steps