Metadata-Version: 2.4
Name: diba-stt
Version: 0.1.0
Summary: Persian speech-to-text by Dibachain — offline, on CPU, real-time.
Author-email: Dibachain <info@dibachain.ir>
License: Apache-2.0
Project-URL: Homepage, https://dibachain.ir
Project-URL: Models, https://huggingface.co/Dibachain/Diba-STT
Project-URL: Demo, https://huggingface.co/spaces/DibaAi/diba-stt
Keywords: persian,farsi,speech-to-text,stt,asr,offline,dibachain,diba
Classifier: Programming Language :: Python :: 3
Classifier: License :: OSI Approved :: Apache Software License
Classifier: Topic :: Multimedia :: Sound/Audio :: Speech
Classifier: Natural Language :: Persian
Requires-Python: >=3.8
Description-Content-Type: text/markdown
License-File: NOTICE
Requires-Dist: vosk>=0.3.44
Requires-Dist: huggingface_hub>=0.20
Provides-Extra: full
Requires-Dist: vosk>=0.3.44; extra == "full"
Dynamic: license-file

# Diba-STT

**Persian speech-to-text by [Dibachain](https://dibachain.ir) — offline, on CPU, real-time.**

Part of the Diba model family. Transcribes Persian audio without a cloud or GPU.

## Install

```bash
pip install diba-stt
```

## Use

```python
from diba_stt import DibaSTT

stt = DibaSTT("fa-small")            # "fa" for the large, more accurate model
print(stt.transcribe("audio.wav"))   # 16 kHz mono WAV
```

Real-time from an audio stream:

```python
stt = DibaSTT("fa-small")
for chunk in pcm_chunks:             # 16-bit PCM bytes
    print(stt.accept(chunk))
print(stt.final())
```

Models: `fa-small` (~53 MB, default), `fa` (~1.6 GB, most accurate), `fa-small-0.5`, `fa-0.5`.
Models are downloaded automatically from [Dibachain/Diba-STT](https://huggingface.co/Dibachain/Diba-STT) on first use.

Live browser demo: https://huggingface.co/spaces/DibaAi/diba-stt

Apache-2.0. See NOTICE for attribution.
