---
license: apache-2.0
base_model: openai/whisper-base.en
library_name: basert
pipeline_tag: automatic-speech-recognition
tags:
  - basert
  - apple-silicon
  - whisper
  - speech-recognition
---

# whisper-base.en

[BaseRT](https://github.com/basecompute/baseRT) `.base` build of
[`openai/whisper-base.en`](https://huggingface.co/openai/whisper-base.en),
OpenAI's 74M-parameter Whisper speech-recognition model (English-only),
for fast local transcription on Apple Silicon.

## Files

| File | Precision | Size |
|------|-----------|------|
| `whisper-base.en-F16.base` | float16 | 149 MB |
| `whisper-base.en-Q8.base` | 8-bit linears, f16 embeddings/conv/norms | 106 MB |
| `whisper-base.en-Q4.base` | 4-bit linears, f16 embeddings/conv/norms | 86 MB |

F16 and Q8 are transcription-quality equivalent (Q8 passes the same
word-error parity gates against reference openai-whisper). Q4 is the
smallest and remains accurate; on some smaller variants it can
occasionally repeat a word in timestamped beam decoding.

## Usage

```bash
curl -LsSf https://basecompute.co/install.sh | sh
basert serve --model whisper-base.en-F16.base
```

`POST /v1/audio/transcriptions` (multipart or JSON) returns `json`, `text`,
`srt`, `vtt`, or `verbose_json` (with per-segment `avg_logprob` /
`no_speech_prob` / `compression_ratio` / `temperature` and the detected
`language`), with optional SSE streaming. Supported
request fields: `language` (or `"auto"` to detect), `prompt`
(initial prompt / vocabulary bias). Or transcribe directly
from the CLI:

```bash
basert-transcribe whisper-base.en-F16.base audio.wav
```

This is the English-only variant; `language` is fixed to `en` and `task=translate` is rejected.

Released under the apache-2.0 license, inherited from the base model.