Downloads · 30 days
0
Reza2kn/Shenava-Koochik-v1.0-sherpa-onnx
Shenava-Koochik-v1.0-sherpa-onnx is a automatic speech recognition model from Reza2kn. Use it when you need speech turned into text. It is set up for sherpa-onnx. The card lists the license as apache-2.0.
Offline Persian ASR bundle for sherpa-onnx. Use this repository when you want sherpa-onnx’s common API across Python, C++, Android, iOS, WebAssembly, C, Go, Swift, Kotlin, or Rust.
Downloads · 30 days
0
Access
Public
Updated Sep 12, 2026
Repo size
459 MB
Likes
5
Public
Click a slice to open those files.
.onnx459 MB · 100%
From the Hugging Face model README
Offline Persian ASR bundle for sherpa-onnx. Use this repository when you want sherpa-onnx’s common API across Python, C++, Android, iOS, WebAssembly, C#, Go, Swift, Kotlin, or Rust.
| English | فارسی | |
|---|---|---|
| 🎯 Role | Cross-platform offline ASR | تشخیص گفتار آفلاین چندسکویی |
| 📦 Bundle | ONNX model, tokens, Persian ITN | مدل ONNX، توکنها و ITN فارسی |
| 🧩 Targets | Desktop, mobile, WebAssembly | دسکتاپ، موبایل و WebAssembly |
| 🎧 Audio | 16 kHz mono | صوت تککانالهٔ ۱۶ کیلوهرتز |
| ⚠️ Key rule | Disable per-feature normalization | نرمالسازی per-feature را خاموش کنید |
Reza2kn/Shenava-Koochik-v1.0-sherpa-onnxPersianML/Shenava-Koochik-v1.0-sherpa-onnxReza2kn/Shenava-Koochik-v1.0model.onnx: offline FastConformer CTC graph.tokens.txt: 1,025-token SentencePiece/CTC vocabulary.persian_itn.py: optional spoken-number to Persian-digit postprocessor.import numpy as np
import sherpa_onnx
import soundfile as sf
from persian_itn import itn
recognizer = sherpa_onnx.OfflineRecognizer.from_nemo_ctc(
model="model.onnx",
tokens="tokens.txt",
num_threads=4,
)
audio, sample_rate = sf.read("speech.wav")
stream = recognizer.create_stream()
stream.accept_waveform(sample_rate, np.asarray(audio, dtype=np.float32))
recognizer.decode_stream(stream)
print(itn(stream.result.text))
Use 16 kHz mono audio. The graph expects unnormalized NeMo log-mel features; its ONNX metadata deliberately leaves normalize_type empty. Enabling sherpa-onnx’s default per-feature normalization produces invalid output for this export.
This is an offline recognizer, not the cache-aware streaming graph. Numbers are emitted in spoken form unless persian_itn.py is applied.
این مخزن بستهٔ آفلاین sherpa-onnx برای تشخیص گفتار فارسی است. نرمالسازی per-feature را فعال نکنید؛ متادیتای مدل عمداً normalize_type را خالی گذاشته است. برای تبدیل عددهای گفتاری به رقم از persian_itn.py استفاده کنید.
🧠 Koochik source · 🌐 Browser ONNX · 🧰 sherpa-onnx · 🍎 Core ML · 🦀 tract offline
Apache-2.0.