{"id":2126,"date":"2026-10-04T13:30:16","date_gmt":"2026-10-04T11:30:16","guid":{"rendered":"https:\/\/www.pcjv.de\/?p=2126"},"modified":"2026-10-04T13:30:16","modified_gmt":"2026-10-04T11:30:16","slug":"go-pocket-tts","status":"publish","type":"post","link":"https:\/\/www.pcjv.de\/?p=2126","title":{"rendered":"Pocket-sized speech, in Go"},"content":{"rendered":"\n

Kyutai’s PocketTTS<\/a> is a text-to-speech model small enough to run on a CPU. Its reference implementation is Python and PyTorch. go-pocket-tts<\/a> runs the same model as a single Go binary: it reads the safetensors weights directly and needs neither Python nor ONNX Runtime.<\/p>\n\n\n\n

It got one sentence in Audio on Linux, in Go<\/a>. It deserves more than that, because upstream has moved on since then: a September model, German, a 24-layer variant and a new kind of sampler. This weekend I brought the port up to upstream v3.3.0. That came to 25 pull requests and 116 commits in two days, and German is part of it.<\/p>\n\n\n\n\n\n\n\n

Listen first<\/h2>\n\n\n\n

Both clips below were made today with the commands shown under them. Each is a single take, cut only by the model’s own end-of-speech decision. The only processing afterwards was loudness normalisation and MP3 encoding.<\/p>\n\n\n\n

English: english_2026-09<\/code> with the default voice alba<\/code>. 8.2 s of audio.<\/figcaption><\/figure>\n\n\n\n
German: the 24-layer german_24l<\/code> with the default voice juergen<\/code>, reading upstream’s own German demo text. 6.1 s of audio.<\/figcaption><\/figure>\n\n\n\n
.\/pockettts synth --language english_2026-09 \\\n  --text \"Hello. I am a text to speech model, running in pure Go on an ordinary laptop CPU. No Python, and no ONNX Runtime.\" \\\n  --out en.wav\n\n.\/pockettts synth --language german_24l \\\n  --text \"Hallo Welt. Ich bin Pocket TTS von Kyutai. Ich bin schnell genug, um auch auf kleinen CPUs zu laufen. Ich hoffe, ich gefalle dir.\" \\\n  --out de.wav<\/code><\/pre>\n\n\n\n
As a check that does not depend on my ears, I ran both files through Whisper. It returned both texts word for word. The only slips were the proper nouns: “Onyx runtime” and “Qtai”.<\/p>\n\n\n\n
Some honest numbers, from a laptop with an AMD Ryzen 5 4600H. The English clip comes from a 6-layer model, and generating it took 10.1 s for 8.2 s of audio, about 0.8\u00d7 real time, with a peak of 1.6 GB of memory. The German clip comes from the 24-layer model, which is about three times the weights: 10.8 s for 6.1 s of audio, about 0.6\u00d7 real time, with a peak of 3.4 GB. So neither is faster than real time on this machine. That is consistent with the model being built to fit a CPU.<\/p>\n\n\n\n
Or try it in your browser<\/h2>\n\n\n\n
No need to build anything to hear it: the web demo<\/a> runs the same Go code, compiled to WebAssembly. Pick a model and a voice, type something and press Synthesize. The page bundles no model of its own. It fetches model, tokenizer and voices straight from Hugging Face, at the same pinned revisions the command line uses.<\/p>\n\n\n\n
\"The
The demo with the 24-layer German model and the voice juergen. The browser is much slower than the native binary: this 6-second clip took 51 s on the same laptop, after a 670 MB download.<\/figcaption><\/figure>\n\n\n\n
Five of the six models work in the browser, german_24l<\/code> included. The sixth, english_2026-09_24l<\/code>, is left out for now. Its checkpoint is 1.3 GB. It loads, but then the Mimi decoder runs out of 32-bit WebAssembly address space with 4.27 GB in use.<\/p>\n\n\n\n
What is in the box<\/h2>\n\n\n\n
The project is about 50,000 lines of Go, and 131 of its 250 source files are tests. It ships two binaries. pockettts<\/code> is the one you use:<\/p>\n\n\n\n