Skip to content Studio Make images and video without living inside the AI toolchain.
A local-first workspace for turning ComfyUI workflows into tools you can use, understand, and reuse.
© 2026 Tea Time Labs
Local AI creation, organized.
Have a ComfyUI workflow? Get a readable Workflow Doc in your browser. Make a workflow readable. Try it now ↗
Ipa Whipstr Base 48khz Cv 21 | ST Studio
Resources / Ipa Whipstr Base 48khz Cv 21 Ipa Whipstr Base 48khz Cv 21 A deep learning automatic speech recognition (ASR) system for transcribing speech audio into IPA text using transformer-based sequence-to-sequence models.
A deep learning automatic speech recognition (ASR) system for transcribing speech audio into IPA text using transformer-based sequence-to-sequence models.
Kind Text model Version checkpoint model
Model source
Kind Text model
File size 179 MB
Version vcheckpoint
Updated Sep 22, 2026
Downloads 14
Source Hugging Face Neurlang Whipstr STT (ASR)
A deep learning automatic speech recognition (ASR) system for transcribing speech audio into IPA text using transformer-based sequence-to-sequence models.
Model artifacts
1 artifact checkpoint.pt
pt · 179 MB · SHA-256 dd7df9a815a5…f9f2 · Hugging Face
Download Source excerpts
2 excerpts
Language: Universal (IPA), 74+ languages
Model Dataset: Common Voice 21
Model-Native Sample Rates: 8000 Hz, 16000 Hz, 24000 Hz, 32000 Hz, 48000 Hz
Degraded-Performance Sample Rates: 11025 Hz, 22050 Hz, 44100 Hz
License: GPL v2
Release: 2026-07-07
Size: 186 MBTotal parameters:
Encoder: 7 220 576Transformer: 7 537 184Total: 14 757 760
CER: 48.68% (51.32% success rate)
Note: Averaged across all supported languages, works better on higher resource languages
WER: 92.58% (7.42% success rate)
Note: Averaged across all supported languages, works better on higher resource languages
Training Details:
Hardware: Nvidia SparkBatch size: 1Samples: 960000Duration: 1:11:10:00Runs: Jul 5 14:58 - Jul 5 17:45, Jul 5 22:41 - Jul 6 04:38, Jul 6 04:46 - Jul 7 07:12 (shut down at Jul 7 07:36)
Inference code git clone https://github.com/neurlang/whipstr.git
cd whipstr/
uv run --with torch --with transformers --with phase-spectrogram stt_infer_hf.py --audio /home/m/Downloads/LJ001-0001.wav --model neurlang/ipa-whipstr-base-48khz-cv-21
Output: Loading weights: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 139/139 [00:00<00:00, 13371.75it/s]
Transcription: ˈɪt ˈæt juː pˈæst ðə kˈɔldz tˈɑɹk fˈeɪs vˈæst ðə ɹˈɛpɪtʃən lˈupənəl ˈi ˈæz bɪhˈeɪviɚ wˈi sˈɑloʊks lˈɛkəmˌɑɹəl hˈæn jˈæt lˈɚnd bˈeɪsɪks sˈɛktɪs ˈækmɪkmɪkəɹɪkənɪkəksɪksɪksɪksɪksɪksɪksɪksɪkstəkstəksɪksɪk
Explaination: IPA Ground truth Comment ˈɪt It Exact match. ˈæt juː gets you /ɡɛts/ was apparently lost, leaving something like "at you." pˈæst past Good match. ðə the Exact. kˈɔldz cold- Reasonable; final /z/ is likely a transcription artifact. tˈɑɹk start The /s/ was dropped and /st/ became /k/, a common recognition error in noisy speech. fˈeɪs phase Very close. vˈæst fast /f/→/v/ substitution. ðə
End of training data: Step 960000 is chosen as this release
Step WER CER 943000 94.21% 49.58% 944000 93.16% 49.96% 945000 94.09% 50.22% 946000 95.25% 50.32% 947000 96.18% 50.30% 948000 93.97% 50.78% 949000 92.93% 49.63% 950000 94.55% 50.84% 951000 94.90% 50.11% 952000 92.35%
---
license: gpl-2.0
language_bcp47:
- hy-AM
- nan-TW
- pa-IN
- sv-SE
- zh-CN
language:
- hy
- nan
- pa
- sv
- zh
- af
- am
- ar
- as
- az
- ba
- be
- bg
- bn
- ca
- cs
- da
- de
- el
- en
- eo
- es
- et
- eu
- fa
- fi
- fr
- gl
- gn
- he
- hi
- ht
- hu
- ia
- id
- is
- it
- ja
- kk
- ko
- ky
- lo
- lt
- ltg
- mk
- ml
- mn
- mr
- mt
- nl
- pl
- ps
- pt
- ro
- ru
- sd
- sk
- sl
- sq
- sw
- ta
- te
- th
- tk
- tn
- tr
- tt
- ug
- uk
- ur
- uz
- vi
- yue
- zu
pipeline_tag: automatic-speech-recognition
tags:
- whipstr
- stt
- asr
---
# Neurlang Whipstr STT (ASR)
A deep learning automatic speech recognition (ASR) system for transcribing speech audio into IPA text using transformer-based sequence-to-sequence models.
- **Language:** Universal (IPA), 74+ languages
- **Model Github:** neurlang/whipstr [https://github.com/neurlang/whipstr](https://github.com/neurlang/whipstr)
- **Model Dataset:** Common Voice 21
- **Model-Native Sample Rates:** 8000 Hz, 16000 Hz, 24000 Hz, 32000 Hz, 48000 Hz
- **Degraded-Performance Sample Rates:** 11025 Hz, 22050 Hz, 44100 Hz
- **License:** GPL v2
- **Release:** 2026-07-07
- **Size:** 186 MB
- **Total parameters:**
- **Encoder:** 7 220 576
- **Transformer:** 7 537 184
- **Total:** 14 757 760
- **CER:** 48.68% (51.32% success rate)
- **Note:** Averaged across all supported languages, works better on higher resource languages
- **WER:** 92.58% (7.42% success rate)
- **Note:** Averaged across all supported languages, works better on higher resource languages
- **Training Details:**
- **Hardware:** Nvidia Spark
- **Batch size:** 1
- **Samples:** 960000
- **Duration:** 1:11:10:00
- **Runs:** Jul 5 14:58 - Jul 5 17:45, Jul 5 22:41 - Jul 6 04:38, Jul 6 04:46 - Jul 7 07:12 (shut down at Jul 7 07:36)
---
## Inference code
```bash
git clone https://github.com/neurlang/whipstr.git
cd whipstr/
uv run --with torch --with transformers --with phase-spectrogram stt_infer_hf.py --audio /home/m/Downloads/LJ001-0001.wav --model neurlang/ipa-whipstr-base-48khz-cv-21
```
### Output:
```bash
Loading weights: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 139/139 [00:00<00:00, 13371.75it/s]
Transcription: ˈɪt ˈæt juː pˈæst ðə kˈɔldz tˈɑɹk fˈeɪs vˈæst ðə ɹˈɛ...
Source context: 14 downloads · 2 likes · Pipeline automatic-speech-recognition · Repo neurlang/ipa-whipstr-base-48khz-cv-21
ɹˈɛpɪtʃən repetition Clearly recognizable.
lˈupənəl loop / no-EOS This portion collapsed badly; "loop no-EOS" became something like "lupənəl."
ˈi ˈæz behavior (beginning omitted)The start of "behavior" appears to have shifted.
bɪhˈeɪviɚ behavior Excellent match.
sˈɑloʊks saw looks This is actually close to "saw looks," with the pause removed.
lˈɛkəmˌɑɹəl like the model Quite distorted, but you can hear the rough rhythm.
hˈæn hadn't The final /t/ and /d/ were lost.
bˈeɪsɪks basic Essentially correct.
sˈɛktɪs seq2seq This was the biggest failure—only the initial /sɛk/ of "seq2seq" survived, while
loops mechanics "mechanics" disappeared entirely.
