loqalkit 1.6.3

Unified language SDK — detection, transcription, romanization, translation, TTS routing

LoqalKit

Unified language SDK — detection, transcription, romanization, translation, and TTS routing.

LoqalKit is the client library for the Loqal language platform. It gives you one consistent API across language detection, speech-to-text, romanization, translation, and text-to-speech, and routes each request to the right engine for the target language.

Install

pip install loqalkit

Optional extras:

pip install "loqalkit[vad]"        # voice-activity gating + segment timestamp refinement
pip install "loqalkit[localize]"   # OOXML document localization
pip install "loqalkit[serve]"      # run the bundled HTTP service

vad is strongly recommended for server use. Without it the voice-activity gate is skipped and the transcription model is fed silence and noise, which it will readily hallucinate whole languages over. Transcription still works without the extra — it is simply less accurate.

Authentication

LoqalKit routes model calls through the Loqal platform, which requires a key. The library raises on first use without one — it performs no useful work standalone.

export MANABU_APP_SECRET="..."

Keys are issued from the Loqal dashboard. Contact [e-mail] for access.

Usage

from pathlib import Path

from loqalkit import (
    detect_language_from_text,
    romanize_batch,
    transcribe,
)

result = transcribe(Path("meeting.wav"), language="he", diarize=True)

detected = detect_language_from_text(result["text"])

roman = romanize_batch(["こんにちは", "ありがとう"], lang_code="ja")

romanize_batch returns a list aligned 1:1 with its input, using None for entries the model did not romanize, so callers can fall back to untransliterated display. It never raises.

Supported Python

Python 3.11 and newer.

License

Proprietary — see LICENSE. Third-party components are listed in THIRD_PARTY_NOTICES.md.