--- title: "Phonon-2.5" description: "Phonon-2.5 downloads, benchmarks and runtimes. A 156 MB open speech recognition model for 25 languages at 11.71 % average word error on FLEURS, for Mac, Linux, Windows and NVIDIA GPUs." canonical: "https://www.fermionresearch.com/models/phonon-2-5/" source: "Fermion Research" --- # Phonon-2.5 Open speech recognition in 25 languages from a 156 MB download. 11.71 % average word error on the FLEURS test sets, ahead of Parakeet Redux at the same size and of the 2.5 GB Parakeet TDT 0.6B v3 it derives from. ## Overview Phonon-2.5 is an open speech recognition model for 25 languages in a 156 MB download. On the FLEURS test sets it averages 11.71 % word error across the 25 languages, against 12.18 % for Parakeet Redux, the other low-bit model of the same size, and 13.26 % for Parakeet TDT 0.6B v3, the 2.5 GB full-precision model it derives from. It is ahead of Redux in 14of the 25 languages, and it holds 101.8 % of the base’s word accuracy averaged over the 25, ahead of the base outright in 14 of them (Greek 20.8 against 36.1, Latvian 14.9 against 23.7, Slovenian 18.5 against 24.7), where a standard post-training 4-bit Parakeet v3 (472 MB) is worse than the base in all 25 at 14.48 %. It is a lower average word error rate across the 25 languages than Whisper large-v3 (20 times its size), Meta MMS-1B-all (25 times), Parakeet TDT 0.6B v3 (16 times) and Whisper large-v3-turbo (10 times); Whisper base, about the same size, makes 4.6 times as many errors (53.71 % against 11.71 %). The same file runs on Apple silicon and on Linux (x86-64 and Arm) and Windows CPUs, with no GPU needed, on NVIDIA GPUs and in the CPU and CUDA Docker images. There is no language setting: it detects the language from the audio. It writes punctuated, capitalized text with a start and end time for every word, and the weights are released under CC-BY-4.0. For English alone, [Phonon-2.5 English](https://www.fermionresearch.com/models/phonon-2-5-english/) is the same size and our most accurate English model. ## Models Phonon-2.5 is a single file. Phonon-2.5 English and Phonon-2 remain available beside it. [The Phonon-2.5 English spec](https://www.fermionresearch.com/models/phonon-2-5-english/)[The Phonon-2 spec](https://www.fermionresearch.com/models/phonon-2/) ## Evaluation All three models were scored on the full FLEURS test set of each of the 25 languages with the same harness and the same text normalisation. Table 1 lists every language; the average over the 25 is 11.71 % for Phonon-2.5, 12.18 % for Parakeet Redux and 13.26 % for Parakeet TDT 0.6B v3. | Language | Phonon-2.5156 MB | Parakeet Redux178 MB | Parakeet TDT 0.6B v32,508 MB | | --- | --- | --- | --- | | English | 6.58 | 7.32 | 6.53 | | Spanish | 4.80 | 4.99 | 4.69 | | French | 8.12 | 9.03 | 6.61 | | German | 6.67 | 7.57 | 5.96 | | Italian | 5.21 | 5.12 | 4.75 | | Portuguese | 6.53 | 6.44 | 6.27 | | Dutch | 9.21 | 9.45 | 7.93 | | Polish | 10.20 | 10.19 | 8.55 | | Russian | 8.67 | 9.01 | 7.21 | | Ukrainian | 7.89 | 8.05 | 7.27 | | Romanian | 11.88 | 11.89 | 13.13 | | Czech | 10.40 | 11.41 | 11.93 | | Slovak | 8.89 | 8.17 | 9.72 | | Bulgarian | 13.34 | 13.02 | 13.48 | | Croatian | 12.24 | 11.43 | 13.21 | | Swedish | 14.53 | 14.70 | 15.44 | | Danish | 18.20 | 17.91 | 19.05 | | Finnish | 12.02 | 11.84 | 13.08 | | Estonian | 11.91 | 10.29 | 14.34 | | Latvian | 14.92 | 14.76 | 23.66 | | Lithuanian | 17.94 | 18.05 | 21.68 | | Hungarian | 15.45 | 15.81 | 15.38 | | Slovenian | 18.53 | 18.88 | 24.70 | | Maltese | 17.91 | 16.02 | 20.86 | | Greek | 20.83 | 33.25 | 36.11 | | Average of 25 | 11.71 | 12.18 | 13.26 | - English Phonon-2.56.58 Parakeet Redux7.32 Parakeet TDT 0.6B v36.53 - Spanish Phonon-2.54.80 Parakeet Redux4.99 Parakeet TDT 0.6B v34.69 - French Phonon-2.58.12 Parakeet Redux9.03 Parakeet TDT 0.6B v36.61 - German Phonon-2.56.67 Parakeet Redux7.57 Parakeet TDT 0.6B v35.96 - Italian Phonon-2.55.21 Parakeet Redux5.12 Parakeet TDT 0.6B v34.75 - Portuguese Phonon-2.56.53 Parakeet Redux6.44 Parakeet TDT 0.6B v36.27 - Dutch Phonon-2.59.21 Parakeet Redux9.45 Parakeet TDT 0.6B v37.93 - Polish Phonon-2.510.20 Parakeet Redux10.19 Parakeet TDT 0.6B v38.55 - Russian Phonon-2.58.67 Parakeet Redux9.01 Parakeet TDT 0.6B v37.21 - Ukrainian Phonon-2.57.89 Parakeet Redux8.05 Parakeet TDT 0.6B v37.27 - Romanian Phonon-2.511.88 Parakeet Redux11.89 Parakeet TDT 0.6B v313.13 - Czech Phonon-2.510.40 Parakeet Redux11.41 Parakeet TDT 0.6B v311.93 - Slovak Phonon-2.58.89 Parakeet Redux8.17 Parakeet TDT 0.6B v39.72 - Bulgarian Phonon-2.513.34 Parakeet Redux13.02 Parakeet TDT 0.6B v313.48 - Croatian Phonon-2.512.24 Parakeet Redux11.43 Parakeet TDT 0.6B v313.21 - Swedish Phonon-2.514.53 Parakeet Redux14.70 Parakeet TDT 0.6B v315.44 - Danish Phonon-2.518.20 Parakeet Redux17.91 Parakeet TDT 0.6B v319.05 - Finnish Phonon-2.512.02 Parakeet Redux11.84 Parakeet TDT 0.6B v313.08 - Estonian Phonon-2.511.91 Parakeet Redux10.29 Parakeet TDT 0.6B v314.34 - Latvian Phonon-2.514.92 Parakeet Redux14.76 Parakeet TDT 0.6B v323.66 - Lithuanian Phonon-2.517.94 Parakeet Redux18.05 Parakeet TDT 0.6B v321.68 - Hungarian Phonon-2.515.45 Parakeet Redux15.81 Parakeet TDT 0.6B v315.38 - Slovenian Phonon-2.518.53 Parakeet Redux18.88 Parakeet TDT 0.6B v324.70 - Maltese Phonon-2.517.91 Parakeet Redux16.02 Parakeet TDT 0.6B v320.86 - Greek Phonon-2.520.83 Parakeet Redux33.25 Parakeet TDT 0.6B v336.11 - Average of 25 Phonon-2.511.71 Parakeet Redux12.18 Parakeet TDT 0.6B v313.26 Table 1Word error rate, %, on the full FLEURS test set of each language, lower is better; the best value in each row is underlined. All three models were scored with the same code and the same text normalisation. Parakeet Redux is a 178 MB low-bit model of the same architecture; Parakeet TDT 0.6B v3 is the 2,508 MB full-precision model both derive from. On the Open ASR Leaderboard’s seven English sets, scored with its code on the full test sets, Phonon-2.5 averages 5.06 % word error, against 5.21 % for Phonon-2 and 4.82 % for Phonon-2.5 English: a better English model than Phonon-2 while covering 24 more languages, from a smaller download. | Model | Download | LS clean | LS other | AMI | Earnings-22 | GigaSpeech | SPGISpeech | VoxPopuli | Average | | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | | Phonon-2.5 English | 158 MB | 1.64 | 3.56 | 8.76 | 6.22 | 7.99 | 3.31 | 2.28 | 4.82 | | Phonon-2.525 languages | 156 MB | 1.60 | 3.41 | 9.12 | 6.72 | 8.04 | 3.31 | 3.23 | 5.06 | | Phonon-2 | 164 MB | 1.72 | 3.92 | 9.37 | 6.96 | 8.35 | 3.70 | 2.46 | 5.21 | | Underdog Bark 3.0 | 163 MB | 1.89 | 4.08 | 7.82 | 7.05 | 7.95 | 3.72 | 2.05 | 4.94 | | Parakeet TDT 0.6B v3† | 2,508 MB | 1.52 | 3.13 | 9.42 | 5.85 | 7.99 | 3.63 | 3.19 | 4.96 | | Parakeet Redux | 178 MB | 1.94 | 4.35 | 9.16 | 7.90 | 8.62 | 4.01 | 3.87 | 5.69 | - Phonon-2.5 English158 MB LS clean1.64 LS other3.56 AMI8.76 Earnings-226.22 GigaSpeech7.99 SPGISpeech3.31 VoxPopuli2.28 Average4.82 - Phonon-2.525 languages156 MB LS clean1.60 LS other3.41 AMI9.12 Earnings-226.72 GigaSpeech8.04 SPGISpeech3.31 VoxPopuli3.23 Average5.06 - Phonon-2164 MB LS clean1.72 LS other3.92 AMI9.37 Earnings-226.96 GigaSpeech8.35 SPGISpeech3.70 VoxPopuli2.46 Average5.21 - Underdog Bark 3.0163 MB LS clean1.89 LS other4.08 AMI7.82 Earnings-227.05 GigaSpeech7.95 SPGISpeech3.72 VoxPopuli2.05 Average4.94 - Parakeet TDT 0.6B v3†2,508 MB LS clean1.52 LS other3.13 AMI9.42 Earnings-225.85 GigaSpeech7.99 SPGISpeech3.63 VoxPopuli3.19 Average4.96 - Parakeet Redux178 MB LS clean1.94 LS other4.35 AMI9.16 Earnings-227.90 GigaSpeech8.62 SPGISpeech4.01 VoxPopuli3.87 Average5.69 Table 2Word error rate, %, lower is better; the best value in each column is underlined. † The Open ASR Leaderboard’s published row; the other rows were scored with its code on the full test sets, with Earnings-22 as the leaderboard’s chunked set. The Phonon-2 row is the one published with its release. ## Speed Phonon-2.5 has the architecture, the file format and the engines of Phonon-2, so it runs at about the same speed as Phonon-2, and it will get a lot faster in the coming weeks. For reference, these are Phonon-2’s published figures, from its release post: 174 times realtime on a base M5 MacBook Air, which is an hour of audio in about 20 seconds; 143 times realtime on eight AMD Zen 5 cores; and 6,680 times realtime batched on an H100. ## Run it Python 3.10 or newer. The fermion command line installs from PyPI; name the model as `phonon-2.5`. ```text $ pip install fermion-research $ fermion transcribe phonon-2.5 meeting.wav $ fermion transcribe phonon-2.5 meeting.wav --json $ fermion serve phonon-2.5 --port 8010 ``` It transcribes files, adds a start and end time for every word under --json, and serves an OpenAI-compatible /v1/audio/transcriptions endpoint. ```text $ pip install "fermion-research[mlx]" ``` On Apple silicon it runs the model on the GPU through MLX. ```text $ pip install --no-deps torch --index-url https://download.pytorch.org/whl/cpu $ pip install fermion-research ``` On Linux (x86-64 and Arm) and Windows the same package runs the CPU engine; on Linux, install torch from its CPU wheel index first to skip the GPU build. [Weights on Hugging Face](https://huggingface.co/FermionResearch/Phonon-2.5)[Documentation](https://www.fermionresearch.com/docs/speech/)[fermion-research on PyPI](https://pypi.org/project/fermion-research/)[Engines on GitHub](https://github.com/fermionresearch/phonon) ## Availability ### License The weights are released under the Creative Commons Attribution 4.0 licence, which Phonon-2.5 inherits from NVIDIA’s Parakeet TDT 0.6B v3; the repository’s NOTICE carries the attribution and lists the changes. The licence permits commercial use, modification and redistribution with attribution. The command line is released under Apache 2.0. ### Citation ```text @misc{fermionresearch2026phonon25, title = {Phonon-2.5}, author = {{Fermion Research}}, year = {2026}, url = {https://fermionresearch.com/models/phonon-2-5/} } ``` ## Related research ![A field of warm mineral light gathering into a single bright band](https://www.fermionresearch.com/images/launch/phonon-hero.jpg) ResearchOctober 22, 2026 ### Introducing Phonon-2.5 Open speech recognition in 25 languages from a 156 MB download, and a new English model of the same size. Both run on a laptop CPU at the speed of Phonon-2. ![A field of warm mineral light gathering into a single bright band](https://www.fermionresearch.com/images/launch/phonon-hero.jpg) ResearchSeptember 29, 2026 ### Introducing Phonon-2 The most accurate open speech recognition model under 900 MB, in a 164 MB download that turns an hour of audio into text in about 20 seconds on a MacBook Air. Source: [https://www.fermionresearch.com/models/phonon-2-5/](https://www.fermionresearch.com/models/phonon-2-5/)