Siddharthanarva/Credresolve-Swar

CredResolve Voice (Swar)

CredResolve Voice is a production-style multilingual TTS application with reusable voice cloning, voice versioning, and a polished web interface.

This repository includes: cr

• a CredResolve-branded web app
• live HTTP audio streaming
• live WebSocket audio streaming
• Prometheus metrics for HTTP, WebSocket, and TTS lifecycle monitoring
• a ready-to-run Grafana and Prometheus monitoring stack
• reusable public voice IDs such as voice_cr_a91f0c2d
• voice version management under the same voice ID
• a developer-facing Integrate page
• a FastAPI backend with typed request handling

Product Overview

CredResolve Voice is built as a full application layer for real-world voice operations. The app is designed for product teams, internal tools, and API-based integrations that need:

• text-to-speech generation
• saved voice cloning
• versioned voice updates
• playback and download
• developer-friendly streaming endpoints

UI Included

The current UI includes these sections:

• Studio

  • text input
  • language selection
  • auto, cloned, and design voice modes
  • male and female voice selection
  • sample rate selection for 24000 Hz and 8000 Hz
  • diffusion controls
  • expression tag shortcuts such as [laughter]
  • streamed audio output with metrics

• Voice Clone

  • reference audio upload
  • optional reference text
  • display name
  • preprocessing toggle
  • reusable saved voice creation

• Voices

  • saved voice library
  • stable public voice IDs
  • active version tracking
  • create new versions under the same voice ID
  • commit a version live for TTS
  • delete old versions safely

• Integrate

  • HTTP stream endpoint
  • WebSocket stream endpoint
  • request field guide
  • cURL example
  • WebSocket URL and first-message example

Voice Registry

Each cloned voice gets a stable public voice ID.

Example:

voice_cr_a91f0c2d

Each voice can have multiple versions. When a new version is committed:

• the voice_id stays the same
• the active version changes
• future TTS requests use the committed version

This makes it possible to improve or replace a saved voice without breaking integrations that already use the public voice ID.

Streaming Behavior

The application supports two streaming paths:

• HTTP

  • POST /api/v1/tts/stream
  • sends a WAV stream in bytes
  • starts with a WAV header
  • then streams audio chunk by chunk as chunks are generated

• WebSocket

  • WS /api/v1/tts/stream/ws
  • sends metadata first
  • then sends binary audio frames progressively
  • ends with a completion message

API Endpoints

Main app endpoints:

• GET /api/v1/capabilities
• GET /api/v1/languages
• GET /api/v1/voices
• POST /api/v1/voices/clone
• POST /api/v1/voices/{voice_id}/versions
• POST /api/v1/voices/{voice_id}/versions/{version_id}/commit
• DELETE /api/v1/voices/{voice_id}
• DELETE /api/v1/voices/{voice_id}/versions/{version_id}
• POST /api/v1/tts/synthesize
• POST /api/v1/tts/stream
• WS /api/v1/tts/stream/ws

Run Locally

Create or activate your virtual environment, install dependencies, and launch the app:

uv sync
source .venv/bin/activate
python3 -m omnivoice.cli.demo --ip 0.0.0.0 --port 8001

You can also use the console script:

credresolve-voice --ip 0.0.0.0 --port 8001

Common Launch Options

python3 -m omnivoice.cli.demo \
  --ip 0.0.0.0 \
  --port 8001 \
  --device cuda \
  --data-dir /home/ubuntu/credresolve_Multi/.credresolve_voice

Useful flags:

• --model
• --device
• --ip
• --port
• --root-path
• --no-asr
• --data-dir

Data Storage

By default, application data is stored under .credresolve_voice/.

This includes:

• saved voice assets
• voice registry metadata
• uploaded reference audio
• generated output WAV files

These runtime artifacts are ignored by Git through .gitignore.

Example HTTP Stream Request

curl -X POST "http://127.0.0.1:8001/api/v1/tts/stream" \
  -H "Content-Type: application/json" \
  --output credresolve-stream.wav \
  -d '{
    "text": "This is a CredResolve Voice streaming request.",
    "language": "auto",
    "voice_mode": "auto",
    "gender": "female",
    "sample_rate": 8000,
    "speed": 1.0,
    "diffusion_steps": 32,
    "guidance_scale": 2.0,
    "denoise": true,
    "preprocess_prompt": true,
    "postprocess_output": true
  }'

Example WebSocket Request

Connect to:

ws://127.0.0.1:8001/api/v1/tts/stream/ws

Then send:

{
  "text": "Streaming over WebSocket",
  "language": "auto",
  "voice_mode": "auto",
  "gender": "auto",
  "sample_rate": 24000
}

Notes

• 8000 Hz output is supported for telephony-style use cases.
• Expression tags are supported directly in text, for example [laughter].
• Saved voices can be selected in Studio or through the API using voice_id.
• Prometheus metrics are exposed at GET /metrics.
• Monitoring setup is documented in docs/monitoring.md.
• The web UI is under omnivoice/app/templates/ and omnivoice/app/static/.
• The FastAPI server is defined in omnivoice/app/server.py.
• The main application service layer is in omnivoice/app/service.py.

Repository Structure

Key files and folders:

• omnivoice/app/server.py
• omnivoice/app/service.py
• omnivoice/app/registry.py
• omnivoice/app/generation.py
• omnivoice/app/schemas.py
• omnivoice/app/templates/index.html
• omnivoice/app/static/app.js
• omnivoice/app/static/styles.css
• omnivoice/cli/demo.py

Status

This repository is currently set up as the CredResolve Voice application codebase, with branding, UI, streaming APIs, saved voice registry support, and voice version management in place.