Voice / speech
Voice — Whisper, XTTS, F5-TTS, voice cloning
ASR and TTS pipelines with a documented install path, not pip install roulette.
Launch standard
Pay by the hour · no contracts · terminate anytime from the dashboard
What this is
Voice pipelines need Whisper for ASR, a good TTS (XTTS-v2, F5-TTS, or Kokoro for speed), and usually a diarization step. Each has its own CUDA quirks. The instance boots on a bare CUDA 12.1 / Ubuntu base — nothing preinstalled — but we publish the working combo below so a few pip installs get you from raw audio to a transcript or a voice clone.
Our value-add
We publish the exact package and CUDA versions that work together for this stack, so you're not debugging library mismatches yourself.
Environment & recommended stack
- Base image: nvidia/cuda 12.1 devel (Ubuntu 22.04)clean boot — no Python ML stack preinstalled
- faster-whisper (Whisper v3 large)recommended — CTranslate2-backed, 2–4× realtime
- WhisperXrecommended for diarization + word-level alignment
- F5-TTS + XTTS-v2recommended for high-quality voice cloning from seconds of audio
- Kokororecommended 82M-param TTS for realtime use cases
- torchaudio + librosa + pyannoterecommended audio plumbing
You stay in control
- Bring your own voice samples — upload via scp or use HF datasets
- Full root to swap in custom ASR models or voice cloning forks
- Prepaid credit cap + auto-destroy deadline for nightly batch transcription jobs
Example workflow
Transcribe + diarize a podcast
# upload the file first
scp episode.mp3 root@$INSTANCE:/root/
ssh root@$INSTANCE
pip install whisperx
whisperx episode.mp3 \
--model large-v3 --diarize \
--hf_token "$HUGGINGFACE_TOKEN" \
--output_format jsonRun inside the instance after `ssh root@<instance-ip>` or via the web terminal.
Launch options
Pick a GPU for voice / speech
Recommended start
Standard — transcription + TTS workflows
48 GB handles Whisper large + F5-TTS / XTTS with batching room
$2.04/ hour
Launch with this configScale option
High throughput — batch transcription
When you need to burn through hundreds of hours of audio fast
$4.41/ hour
Launch with this configHourly rates shown are current target rates — the authoritative price is surfaced on the /launch page at the moment you provision. Billing is per 5 minutes of runtime.