Little TTS
A 4.65M-parameter English text-to-speech model with two voices, trained from scratch. Speech runs locally in your browser, without PyTorch, quantization, or a GPU. Your text stays on your device. Noncommercial use only.
Voice 92 is recommended. These examples favour narrative reading; pronunciation and naturalness can still vary.
- Audio
- –
- Time to generate
- –
- Speed
- –
- Real-time factor
- –
Phonemes
CPU speed and memory
Native ONNX Runtime benchmark: Apple M4 Pro, two threads, 18 utterances. RTF 0.0091 (110× real time), peak RAM 502 MiB. These are native Python measurements, not browser speed or RAM. Your browser's measured generation speed appears above after synthesis.
Model: 19.5 MB, plus about 5.4 MB pronunciation files and browser runtimes. No GPU or PyTorch is required. English only, up to 350 characters per request. Longer inputs and mobile browsers may need more time and memory.