MetaVoiceUnclaimed profile

Conversational speech models for real-time voice agents — human-sounding phone calls with sub-250ms time-to-first-audio.

Task: Composing Music & AudioUse case: Build and deploy voice agents for sales lead qualification and customer conversaModels: speech-1

0.0

based on 0 reviews

Are you the vendor?

reviews

Manzia generates an at-a-glance sentiment summary once an Agent has 5 approved reviews. This Agent has 0.

Summarized from 2 community tutorials · July 2026

Access: Free trial
Runs on: Self-hosted · CLI · API
You’ll need: NVIDIA GPU (A100 recommended; T4 had issues)
Python 3.11 + conda environment
~30s reference audio clip of target voice
Git/GitHub repo clone
Skill level: Developer — code or a terminal

Summarized from 2 community tutorials · July 2026

Zero-shot voice cloning from ~30s audioClones a target voice (e.g. Zuckerberg, Lex Fridman, or the creator's own recorded clip) from just a short reference audio sample, no fine-tuning needed.
Self-hosted TTS server with API endpointRuns fam_llm serving.py to stand up a local server exposing a /tts endpoint that accepts text + a reference voice file and returns cloned speech.
Temperature / guidance-scale tuningAdjusting temperature and guidance scale changes accent, pacing and how closely the output matches the reference voice.
Command-line sample generationfam_llm/sample.py generates test audio from a built-in voice and text prompt straight from the terminal for quick verification.
Runs on your own GPU, no vendor account neededModel weights auto-download from Hugging Face and inference runs entirely on a local or Colab GPU rather than a hosted MetaVoice product.

Summarized from 2 community tutorials · July 2026