Open 164 MB English speech recognition model from Fermion Research, derived from Parakeet TDT 0.6B v3.
Updated EU status checked Last shown
Sources: the vendor's own pages (linked above) and creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.
Phonon-2 is an open English speech-to-text model from Fermion Research. It is a compressed derivative of NVIDIA's Parakeet TDT 0.6B v3: the encoder stores each weight as one of five learned levels (about 2.1 bits per weight), giving a 164 MB download compared with 2.5 GB for the full-precision teacher.\n\nFermion reports an average word error rate of 5.21% across the seven English sets of the Open ASR Leaderboard, against 4.96% for the teacher. It is reported to beat the teacher on meetings (AMI) and parliamentary speech (VoxPopuli), and to stay ahead of Parakeet Redux, another low-bit model of similar size, under added noise. These figures are vendor-reported.\n\nThe same file runs on Apple silicon (via MLX on the GPU), x86-64 and Arm CPUs, Windows, Linux and NVIDIA GPUs. Fermion reports 174× realtime on an M5 MacBook Air (about 20 seconds for an hour of audio), 142.8× on eight Zen 5 cores, and 267× to 3,614× on an A100 depending on batch size. It supports English only. The weights are released under CC-BY-4.0, inherited from the Parakeet base model.
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
An open-source speech-to-text model under 900 MB that transcribes an hour of audio in about 20 seconds. It has the second-lowest average error rate versus much larger models.
Closest neighbours by meaning, not just by category.
Model
Open-source append-only streaming speech recognition model with configurable low-latency decoding, English and Mandarin.
Model · Edge0
Open-weights 0.17B-parameter multilingual text-to-speech model with zero-shot voice cloning, runs on CPU via ONNX.
Model
Speech recognition model trained for noisy, reverberant and distorted real-world audio, with RL post-training.
Model
8.5 MB, 8M-parameter text-to-speech model distilled from Kokoro-82M, with four preset voices and CPU inference.
Model · Nyra Labs
Controllable multilingual speech recognition with verbatim or clean output and word-level timestamps, from Nyra Labs.
Model · Boson AI
Text-to-speech model from Boson AI with inline tags for emotion, style, speed, pitch, pauses and sound effects.
