BREAKING
Sopro TTS Released: 135M Params
0
M
parameters
0
x
RTF on CPU
0
ms
time to first audio
Sopro vs Heavy TTS Models
Sopro
lightweight
●
WaveNet-style convolutions
●
Runs on CPU, no GPU
●
135M parameters
XTTS v2 / VoiceCraft
heavy
●
Transformer-based
●
Large compute demand
●
High quality
Zero-Shot Voice Cloning
1
Record 3-12s clip
↓
2
Feed reference
↓
3
Clone voice
↓
4
Stream output
Cheap to Train, Quality Uneven
Lightweight Local Voice Option
AI NEWS BLITZ
A new open-source lightweight text-to-speech model called Sopro has just launched.