# CrisperWhisper – Speech to Text Model That Transcribes What You Say Verbatim

> Source: <https://github.com/nyrahealth/CrisperWhisper>
> Published: 2026-09-06 23:41:11+00:00

**The most accurate verbatim speech recognition you can run in production:
controllable, multilingual, and timed to the word.**

[Release post](https://www.nyra-labs.com/crisperwhisper) ·
[Paper](https://arxiv.org/abs/2607.18934) ·
[Full documentation](/nyrahealth/CrisperWhisper/blob/main/DOCS.md) ·
[Models](https://huggingface.co/nyralabs) ·
[Benchmark](https://www.nyra-labs.com/research/nyra-verbatim-speech-benchmark)

Most speech-to-text systems never actually decide whether to write down what
was *said* or what was *meant*. They inherit that choice from their training
data and apply it inconsistently. CrisperWhisper 2.0 makes it an explicit,
controllable choice. One recording, two transcripts:

**Verbatim**, exactly what was said, in one consistent format:
`[um] so we we need to, to reschedule the th- thursday meeting to [uh] march third at nine thirty [laughter]`

**Intended**, the clean version the speaker meant, with numbers, dates,
and emails formatted the way you'd write them:
`So we need to reschedule the Thursday meeting to March 3 at 9:30.`

On top of that:

- **Word-level timings.** Around 30 ms mean boundary error on read speech
and 41 ms on conversational speech, the most precise word timing of any
system we benchmarked, on both.
- **Verbatimize.** Upgrade transcripts you already have: given audio plus a
trusted clean transcript, the model reproduces your content word-for-word
and inserts only the disfluencies and vocal events actually present in the
audio (rare-word recall jumps from 6.8% to 96.1% vs. re-transcribing).
This turns the world's abundant clean corpora into verbatim ones, ready
for TTS data, clinical speech analysis, and dataset construction.
- **Multilingual.** Verbatim and intended modes work across most languages
Whisper supports. CrisperWhisper 2.0 tops the[Nyra Verbatim Speech Benchmark](https://www.nyra-labs.com/research/nyra-verbatim-speech-benchmark) leaderboard for disfluency F1 across ten languages, ahead of every
closed-source alternative we tested.
- **Seamless longform.** Audio of any length, transcribed without the usual
chunk-boundary artifacts: each window continues from the words already
transcribed (*conditional continuation* ), so there are no duplicated or
dropped words at the seams and no fragile timestamp-token bookkeeping.
- **Production inference.** A CTranslate2 runtime with speculative decoding
and built-in mitigation of Whisper's looping-hallucination failure mode.

The [Nyra Verbatim Speech Benchmark](https://www.nyra-labs.com/research/nyra-verbatim-speech-benchmark)
scores fillers, repetitions, cut-offs, and vocal sounds as separate, typed
metrics. Its headline number is **disfluency F1**: how reliably a system
writes down the disfluencies that were actually spoken, without inventing
ones that weren't. Averaged over ten languages:

| # | System | Disfluency F1 | 
|---|---|---|
| 1 | **CrisperWhisper 2.0 Pro** | **93.5** | 
| 2 | **CrisperWhisper 2.0** | **87.8** | 
| 3 | ElevenLabs Scribe v2 | 79.2 | 
| 4 | Microsoft MAI-Transcribe-1.5 | 77.5 | 
| 5 | CrisperWhisper 1.0* | 64.8 | 
| 6 | Inworld STT | 59.5 | 
| 7 | xAI Grok Speech-to-Text | 42.8 | 
| 8 | Deepgram Nova-3 | 37.8 | 
| 9 | Fish Audio ASR | 35.0 | 
| 10 | AssemblyAI Universal-3 Pro | 30.5 | 

<sub>* CrisperWhisper 1.0 is English/German-only; its average covers those
two languages. English and German use human-labeled evaluation sets; the
other eight languages use synthetic verbatim sets. Per-language breakdowns
and how the metric is computed are in the
[benchmark post](https://www.nyra-labs.com/research/nyra-verbatim-speech-benchmark).</sub>

Mean absolute word-boundary error on read speech (TIMIT), lower is better:

| # | System | Boundary error | 
|---|---|---|
| 1 | **CrisperWhisper 2.0** | **29.6 ms** | 
| 2 | xAI Grok Speech-to-Text | 37.1 ms | 
| 3 | CTC-seg | 49.3 ms | 
| 4 | ElevenLabs Scribe v2 | 51.3 ms | 
| 5 | NeMo-FA | 60.0 ms | 
| 6 | Deepgram Nova-3 | 63.3 ms | 
| 7 | WhisperX | 64.8 ms | 
| 8 | Cartesia Ink-Whisper | 69.4 ms | 
| 9 | Canary | 85.5 ms | 

<sub>Scored on exactly the words each system gets right. How the timings
are extracted from supervised cross-attention, plus results on
conversational speech, are in the
[aligner post](https://www.nyra-labs.com/research/attention-to-aligner).</sub>

```
# NVIDIA GPU (Linux): fastest, includes speculative decoding.
# An NVIDIA driver is all you need; CUDA libraries arrive via pip.
pip install "crisperwhisper[ct2]"

# Pure PyTorch: runs anywhere torch does (macOS, Windows, CPU)
pip install "crisperwhisper[transformers]"
python
from crisperwhisper import CrisperWhisperModel

model = CrisperWhisperModel()          # nyralabs/CrisperWhisper2.0_large
# or pick a size: CrisperWhisperModel("turbo")  # turbo / medium / small

# Verbatim transcription (default): every filler, repetition, stutter,
# false start, and vocal event
result = model.transcribe("meeting.wav", language="en")
print(result.text)

# Intended: the clean, readable version
clean = model.transcribe("meeting.wav", language="en", mode="intended")

# Word-level timestamps
result = model.transcribe("meeting.wav", language="en", word_timestamps=True)
for w in result.words:
    print(f"{w.start:6.2f}-{w.end:6.2f}  {w.word}")

# Verbatimize: upgrade an existing clean transcript with the
# disfluencies that are actually in the audio
result = model.verbatimize("clip.wav", "I think we should ship it Friday.")
```

Audio longer than 30 seconds is handled automatically (see
[longform](#what-else-is-in-the-box) below). The first load of a model
downloads it from HuggingFace and, on the `ct2` backend, converts it once
into a local cache. That **one-time conversion** needs torch and
transformers, which the lean `[ct2]` extra does not install -- add them
once with:

```
pip install "crisperwhisper[ct2,convert]"   # or add [convert] later
```

A CPU-only torch is enough for the conversion, and the dependencies can
be uninstalled once the converted model is cached. Loading a directory
that already contains a converted CT2 model (`model.bin`) never needs
them.

| Shorthand | HuggingFace ID | Notes | 
|---|---|---|
| `"large"` (default) | `nyralabs/CrisperWhisper2.0_large` | Best open quality | 
| `"turbo"` | `nyralabs/CrisperWhisper2.0_turbo` | Fastest, with some quality degradation; recommended as the speculative draft | 
| `"medium"` | `nyralabs/CrisperWhisper2.0_medium` | Near-large quality; best tradeoff between size and quality | 
| `"small"` | `nyralabs/CrisperWhisper2.0_small` | Smallest | 
| `"large_pro"` /`"turbo_pro"` /`"medium_pro"` /`"small_pro"` | `nyralabs/CrisperWhisper2.0_<size>_pro` | **Pro** : our best models, with improved performance, hotword boosting, trained on additional proprietary data | 

The standard models are released under a
[non-commercial research license](https://huggingface.co/nyralabs/CrisperWhisper2.0_large/blob/main/LICENSE.md)
and are available for commercial licensing. The **Pro** models are available
under commercial license only. For both,
[get in touch](https://www.nyra-labs.com/crisperwhisper).

A small draft model proposes tokens and the main model verifies them. Same output, 1.3 to 1.4x faster:

```
model = CrisperWhisperModel("large", draft_model="turbo")
result = model.transcribe("meeting.wav", language="en",
                          speculative_decoding=True)
```

Everything below works out of the box and is covered in depth in
[DOCS.md](/nyrahealth/CrisperWhisper/blob/main/DOCS.md):

| Option | What it does | 
|---|---|
| `mode="verbatim" / "intended"` | Choose what-was-said vs. what-was-meant per call | 
| `word_timestamps=True` | Per-word start/end times from supervised cross-attention alignment | 
| `hotwords=[...]` | Bias recognition toward names and rare terms ( **Pro models only** ) | 
| `model.transcribe_dual(...)` | Verbatim **and** intended in one pass (ct2) | 
| `model.verbatimize(audio, transcript)` | Insert real disfluencies into a trusted clean transcript | 
| `model.forced_align(audio, text)` | Timings for a transcript you already have | 
| Longform | Audio >30s transcribed seamlessly via conditional continuation, with no chunk-boundary duplicates, drops, or stitching | 
| Hallucination mitigation | On by default: detects and suppresses Whisper's looping-repetition failure mode during decoding | 
| `compute_type="float16" / "int8_float16"` | Quantization | 

Each mechanism has a deep-dive post:

- [Measuring verbatimness: the Nyra Verbatim Speech Benchmark](https://www.nyra-labs.com/research/nyra-verbatim-speech-benchmark) .
Typed metrics for fillers, repetitions, cut-offs, and vocal sounds instead
of one opaque WER number.
- [How we unlocked multilingual style-controlled transcription at scale](https://www.nyra-labs.com/research/multilingual-style-control) .
Verbatim/intended control across languages.
- [Turning emergent cross-attention into a precise aligner](https://www.nyra-labs.com/research/attention-to-aligner) .
Supervising Whisper's alignment heads for word boundaries around 30 ms.
- [Longform transcription with conditional continuation](https://www.nyra-labs.com/research/longform-continuation) .
Resolving window seams with the words already transcribed instead of
fragile timestamp tokens.
- [Closing the verbatim data gap with Verbatimize](https://www.nyra-labs.com/research/verbatimize) .
Upgrading clean corpora into verbatim ones at scale.
- [Faster inference and mitigating hallucinations](https://www.nyra-labs.com/research/killing-hallucinations) .
The CTranslate2 stack, speculative decoding, and the anti-looping decoder.

[DOCS.md](/nyrahealth/CrisperWhisper/blob/main/DOCS.md) covers the full API: backends and their trade-offs,
every `transcribe()` option, dual-mode transcription, forced alignment,
longform strategies, speculative-K tuning, hallucination-repair thresholds,
quantization, model conversion, and the result object.

**The inference code in this repository is MIT-licensed** (see
[LICENSE](/nyrahealth/CrisperWhisper/blob/main/LICENSE)): use it freely, commercially or otherwise.
`crisperwhisper/features.py` is vendored from
[faster-whisper](https://github.com/SYSTRAN/faster-whisper) (MIT, SYSTRAN).

**The model weights are not MIT.** They are released under the
[Nyra Health Non-Commercial Research License](https://huggingface.co/nyralabs/CrisperWhisper2.0_large/blob/main/LICENSE.md):
free for research and other non-commercial use; any commercial use requires
a commercial license. The Pro models are available under commercial license
only. For commercial licensing of either,
[contact Nyra](https://www.nyra-labs.com/crisperwhisper).