eworkerinc/vibevoice repository overview

nvidia-smi
docker run --rm --gpus all nvidia/cuda:12.4.1-base-ubuntu24.04 nvidia-smi

docker pull eworkerinc/vibevoice:latest   # alias of :full (prebaked)

sudo mkdir -p /mnt/vv-hf /mnt/vv-voices /mnt/vv-state \
  && sudo chown -R "$USER":"$USER" /mnt/vv-hf /mnt/vv-voices /mnt/vv-state

docker run -d --name vibevoice --gpus all \
  -p 8745:8745 \
  -v /mnt/vv-hf:/root/.cache/huggingface \
  -e ENABLE_1_5B=true -e ENABLE_LARGE=true \
  -e AUTH_REQUIRED=true \
  -e CORS_ENABLED=true -e ALLOWED_ORIGINS='*' \
  -v /mnt/vv-voices:/app/voices \
  -v /mnt/vv-state:/var/lib/eworker \
  eworkerinc/vibevoice:latest

docker run -d --name vibevoice-1_5b --gpus all \
  -p 8745:8745 -v /mnt/vv-hf:/root/.cache/huggingface \
  -e ENABLE_1_5B=true -e ENABLE_LARGE=false \
  eworkerinc/vibevoice:latest

docker run -d --name vibevoice-large --gpus all \
  -p 8745:8745 -v /mnt/vv-hf:/root/.cache/huggingface \
  -e ENABLE_1_5B=false -e ENABLE_LARGE=true \
  eworkerinc/vibevoice:latest

docker run -d --name vibevoice \
  -p 80:80 -p 443:443 \
  -v /mnt/vv-hf:/root/.cache/huggingface \
  -v /mnt/vv-state:/var/lib/eworker \
  -e TLS_MODE=acme -e TLS_DOMAIN=voice.example.com -e [email protected] \
  -e AUTH_REQUIRED=true -e ENABLE_1_5B=true -e ENABLE_LARGE=true \
  eworkerinc/vibevoice:latest

curl -s http://localhost:8745/health | jq .

KEY=$(docker logs vibevoice 2>&1 | sed -n 's/^X-API-Key: //p' | tail -1)

curl -s http://localhost:8745/v1/voice/voices -H "X-API-Key: $KEY" | jq
# If only Large is enabled, include the model filter:
curl -s "http://localhost:8745/v1/voice/voices?model=vibevoice-large" -H "X-API-Key: $KEY" | jq

curl -s "http://localhost:8745/v1/voice/voices/Alice/preview?text=Hello" \
  -H "X-API-Key: $KEY" --output preview.wav

cat > body.json <<'JSON'
{
  "model": "vibevoice-1.5b",
  "script": "Speaker 1: Hello there!\nSpeaker 2: Hi! Great to meet you.",
  "speakers": [ { "voiceName": "Alice" }, { "voiceName": "Carter" } ],
  "overrides": {
    "guidance": { "inference_steps": 28, "cfg_scale": 4.5 }
  }
}
JSON

JOB_ID=$(curl -s -X POST http://localhost:8745/v1/voice/jobs \
  -H "Content-Type: application/json" -H "X-API-Key: $KEY" \
  --data-binary @body.json | jq -r .job_id)

curl -s "http://localhost:8745/v1/voice/jobs/$JOB_ID/result" -H "X-API-Key: $KEY" \
  | jq -r .audio_wav_base64 | base64 --decode > out.wav

# 1) Place your consented voice sample WAV on the host (3–10 seconds is ideal)
#    Requirements: mono PCM, 16‑bit, 16k or 24k sample rate recommended.

# 2) Copy or move it into the mounted voices folder (host side)
cp ~/Downloads/my-voice.wav /mnt/vv-voices/

# 3) List voices (the filename without .wav becomes the voice name)
curl -s http://localhost:8745/v1/voice/voices -H "X-API-Key: $KEY" | jq

# 4) Preview the custom voice (replace MyVoice with your filename stem)
curl -s "http://localhost:8745/v1/voice/voices/MyVoice/preview?text=Hello%20from%20my%20custom%20voice" \
  -H "X-API-Key: $KEY" --output custom-preview.wav

# 5) Use it in a job
cat > job-custom.json <<'JSON'
{
  "model": "vibevoice-1.5b",
  "script": "Speaker 1: This is my custom narrator.",
  "speakers": [ { "voiceName": "MyVoice" } ],
  "overrides": { "guidance": { "inference_steps": 24, "cfg_scale": 4.2 } }
}
JSON
JOB_ID=$(curl -s -X POST http://localhost:8745/v1/voice/jobs \
  -H "Content-Type: application/json" -H "X-API-Key: $KEY" \
  --data-binary @job-custom.json | jq -r .job_id)
curl -s "http://localhost:8745/v1/voice/jobs/$JOB_ID/result" -H "X-API-Key: $KEY" \
  | jq -r .audio_wav_base64 | base64 --decode > custom-out.wav

cat > tiny.json <<'JSON'
{
  "model": "vibevoice-1.5b",
  "script": "Speaker 1: Hello from VibeVoice.",
  "speakers": [ { "voiceName": "Alice" } ],
  "overrides": { "guidance": { "inference_steps": 20, "cfg_scale": 4.0 } }
}
JSON

JOB_ID=$(curl -s -X POST http://localhost:8745/v1/voice/jobs \
  -H "Content-Type: application/json" -H "X-API-Key: $KEY" \
  --data-binary @tiny.json | jq -r .job_id)

curl -s "http://localhost:8745/v1/voice/jobs/$JOB_ID/result" -H "X-API-Key: $KEY" \
  | jq -r .audio_wav_base64 | base64 --decode > tiny.wav

cat > tiny2.json <<'JSON'
{
  "model": "vibevoice-1.5b",
  "script": "Speaker 1: Good morning!\nSpeaker 2: Morning—ready to start?",
  "speakers": [ { "voiceName": "Mary" }, { "voiceName": "Frank" } ],
  "overrides": { "guidance": { "inference_steps": 22, "cfg_scale": 4.2 } }
}
JSON

JOB_ID=$(curl -s -X POST http://localhost:8745/v1/voice/jobs \
  -H "Content-Type: application/json" -H "X-API-Key: $KEY" \
  --data-binary @tiny2.json | jq -r .job_id)

curl -s "http://localhost:8745/v1/voice/jobs/$JOB_ID/result" -H "X-API-Key: $KEY" \
  | jq -r .audio_wav_base64 | base64 --decode > tiny2.wav

cat > roundtable_en.json <<'JSON'
{
  "model": "vibevoice-1.5b",
  "script": "Speaker 1: Welcome everyone to our quick roundtable.\nSpeaker 2: Thanks—let's share highlights from this week.\nSpeaker 3: I experimented with new prompt flows and got cleaner outputs.\nSpeaker 4: Nice. I focused on latency and shaved a few seconds off.\nSpeaker 1: Did the new sampler help with stability?\nSpeaker 2: Yes, especially around tricky words and punctuation.\nSpeaker 3: I also tried a different CFG scale for warmer tone.\nSpeaker 4: And fewer inference steps were fine for short clips.\nSpeaker 1: What about longer narrations?\nSpeaker 2: We should raise steps slightly to keep quality.\nSpeaker 3: Agreed—thirty two felt like a sweet spot.\nSpeaker 4: We could A/B test with audience feedback.\nSpeaker 1: Let's line up a small user study next week.\nSpeaker 2: I can prepare scripts and scenarios.\nSpeaker 3: I will track timing and perceived naturalness.\nSpeaker 4: I'll measure CPU and GPU utilization.\nSpeaker 1: Great. Any blockers we should address?\nSpeaker 2: None on my side.\nSpeaker 3: All good here.\nSpeaker 4: Same—excited to ship improvements.\nSpeaker 1: Perfect—thanks everyone for the quick sync.",
  "speakers": [
    { "voiceName": "Alice" },
    { "voiceName": "Carter" },
    { "voiceName": "Frank" },
    { "voiceName": "Mary" }
  ],
  "overrides": { "guidance": { "inference_steps": 28, "cfg_scale": 4.5 } }
}
JSON

JOB_ID=$(curl -s -X POST http://localhost:8745/v1/voice/jobs \
  -H "Content-Type: application/json" -H "X-API-Key: $KEY" \
  --data-binary @roundtable_en.json | jq -r .job_id)

curl -s "http://localhost:8745/v1/voice/jobs/$JOB_ID/result" -H "X-API-Key: $KEY" \
  | jq -r .audio_wav_base64 | base64 --decode > roundtable_en.wav

cat > bilingual.json <<'JSON'
{
  "model": "vibevoice-1.5b",
  "script": "Speaker 1: Hello and welcome to our bilingual demo.\nSpeaker 2: 大家好，欢迎来到我们的双语演示。\nSpeaker 1: We'll alternate between English and Chinese to compare styles.\nSpeaker 2: 我会用较自然的语气来朗读示例句子。\nSpeaker 1: First, a short introduction to the topic.\nSpeaker 2: 接下来，我们会用简短的段落来说明要点。\nSpeaker 1: Pay attention to clarity and pacing across languages.\nSpeaker 2: 请留意语速与停顿是否听起来自然。\nSpeaker 1: Finally, we'll wrap up with a brief summary.\nSpeaker 2: 最后，我们会做一个简短的总结。\nSpeaker 1: Thanks for listening.\nSpeaker 2: 谢谢收听。",
  "speakers": [ { "voiceName": "Alice" }, { "voiceName": "Xinran" } ],
  "overrides": { "guidance": { "inference_steps": 28, "cfg_scale": 4.5 } }
}
JSON

JOB_ID=$(curl -s -X POST http://localhost:8745/v1/voice/jobs \
  -H "Content-Type: application/json" -H "X-API-Key: $KEY" \
  --data-binary @bilingual.json | jq -r .job_id)

curl -s "http://localhost:8745/v1/voice/jobs/$JOB_ID/result" -H "X-API-Key: $KEY" \
  | jq -r .audio_wav_base64 | base64 --decode > bilingual.wav

docker run --rm eworkerinc/vibevoice:latest bash -lc \
  'ls -1 /root/.cache/huggingface/hub | grep models--microsoft--VibeVoice'