Text Input
G2P Output — Phoneme Sequence
waiting for input...
Mel Spectrogram · 80 bands · 24kHz
FRAMES —
DURATION —
VOCODER NEURALHIFI-GAN
Voice
Synth
★ Studio
Model
Log
Voice Space · 10-dim
measured baritone, warm resonance
Synthesis Parameters
RATE
1.00x
PITCH SHIFT
±0
GL ITERS
24
MVC LAYERS
2
★ Voice Capture
ready — mic not yet requested
Analyze + Train
waiting for recording...
Spoken text (for training)
Required for meaningful duration training.
Training adapts voice timbre and speaking tempo to your recording.
The duration model learns your natural phoneme timing — exported in the .pop2 file.
More recordings = better generalisation.
Voice Identity (.pop2)
no voice trained yet
Model Status
Backbone MVC Bidirectional Mamba
Vocoder HiFi-GAN BitLinear
Weights BitLinear ternary
D_model 128
N_layers 2
Voc stages3
N_mels 80
Sample rate24000 Hz
Hop size 256
N_fft 1024
GL fallbackstandby
Pop2 not loaded
postMessage API
// embed in iframe, control from host:
speak · stop · getStatus
getVoiceVec · setVoiceVec
loadPop2 · loadPop2Url
exportPop2 · reset
// example:
iframe.contentWindow
.postMessage({
target:'omnivocal',
action:'speak',
text:'hello'
}, '*');
System Log