viet-asr 0.1.0.dev5

Universal Vietnamese Speech AI SDK

vietasr — Python binding

Offline Vietnamese Speech AI SDK for Python. Built on top of the same C/C++ core that powers every other vietasr binding.

Install

pip install vietasr

The wheel ships a prebuilt libvietasr.{so,dylib,dll} for your platform under vietasr/_native/. No system dependencies.

Quickstart

import vietasr

pipe = vietasr.Pipeline.preset("transcribe")
result = pipe.transcribe("audio.wav")
print(result.text)

Streaming

import vietasr
import numpy as np

pipe = vietasr.Pipeline.preset("transcribe")
with pipe.stream(sample_rate=16000) as session:
    for chunk in mic_chunks():
        session.accept(np.asarray(chunk, dtype=np.int16))
        print(session.partial().text)
    print("FINAL:", session.final().text)

Custom pipeline

pipe = (vietasr.Pipeline.new()
    .add("vad")
    .add("vietasr")
    .add("punctuation")
    .add("itn")
    .add("gender")
    .add("emotion")
    .build())

result = pipe.transcribe("call.wav")
print(result.text, result["gender"], result["emotion"])

CLI

The wheel installs a vietasr command:

vietasr audio.wav
vietasr --preset analytics call.wav --pretty
vietasr --module vad --module vietasr --module gender audio.wav

Build from source

git clone https://github.com/dangvansam/viet-asr
cd viet-asr/bindings/python
pip install -e .

scikit-build-core drives the CMake build under the hood. The compiled libvietasr is placed in vietasr/_native/ and loaded by the CFFI wrapper.

API reference

Object	Method	Returns
Pipeline	preset(name)	Pipeline
Pipeline	new()	Pipeline
Pipeline	add(module, config=None)	Pipeline (self)
Pipeline	set_backend(backend)	Pipeline (self)
Pipeline	set_model_dir(path)	Pipeline (self)
Pipeline	build()	Pipeline (self)
Pipeline	transcribe(source, sample_rate=16000)	Result
Pipeline	stream(sample_rate=16000)	Session (context manager)
Session	accept(pcm)	bool
Session	partial() / result() / final()	Result
Session	reset()	None
Result	.text, .partial, .is_final, .segments, .speakers	typed
Result	.field(key) / result[key]	Any
Result	.to_json()	str