maqsudxo1ja

uz-whisper-small-stt-v2

README

License: apache-2.0

📊 Natijalar

Table with columns: Ko'rsatkich, Qiymat
Ko'rsatkich	Qiymat
WER (Word Error Rate)	28.64%
Validation Loss	0.2634
Model hajmi	~242M parametr (Whisper-small)

Table with columns: Training Loss, Epoch, Step, Validation Loss, WER
Training Loss	Epoch	Step	Validation Loss	WER
3.0847	0.68	400	0.3161	32.49
2.4038	1.35	800	0.2827	29.58
2.0104	2.03	1200	0.2708	30.53
1.9056	2.70	1600	0.2647	28.92
1.9695	3.00	1776	0.2634	28.64

🚀 Tezkor boshlash

O'rnatish

bash
pip install transformers torch librosa soundfile

1-usul: `pipeline` bilan (eng oson)

python
from transformers import pipeline

transcriber = pipeline(
    "automatic-speech-recognition",
    model="maqsudxo1ja/uz-whisper-small-stt-v2"
)

result = transcriber("audio_fayl.mp3")
print(result["text"])

2-usul: to'g'ridan-to'g'ri model bilan (batafsil nazorat)

python
import torch
import librosa
from transformers import WhisperProcessor, WhisperForConditionalGeneration

MODEL_ID = "maqsudxo1ja/uz-whisper-small-stt-v2"

processor = WhisperProcessor.from_pretrained(MODEL_ID, language="uzbek", task="transcribe")
model = WhisperForConditionalGeneration.from_pretrained(MODEL_ID)

device = "cuda" if torch.cuda.is_available() else "cpu"
model = model.to(device)

# Audio faylni 16kHz formatida yuklaymiz (Whisper shuni talab qiladi)
audio, sr = librosa.load("audio_fayl.mp3", sr=16000)

input_features = processor(
    audio, sampling_rate=sr, return_tensors="pt"
).input_features.to(device)

predicted_ids = model.generate(input_features, language="uzbek", task="transcribe")
transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)[0]

print(transcription)

3-usul: mikrofondan real vaqtda (namuna)

python
import sounddevice as sd
import numpy as np
from transformers import pipeline

transcriber = pipeline("automatic-speech-recognition", model="maqsudxo1ja/uz-whisper-small-stt-v2")

duration = 5  # soniya
print("🎙️  Gapiring...")
audio = sd.rec(int(duration * 16000), samplerate=16000, channels=1, dtype="float32")
sd.wait()

result = transcriber({"array": audio.flatten(), "sampling_rate": 16000})
print("Natija:", result["text"])

🎯 Qo'llanish sohalari

O'zbek tilidagi audio/video kontentni matnga aylantirish
Ovozli buyruqlar va ovozli yordamchilar
Suhbat/qo'ng'iroqlarni avtomatik transkripsiya qilish
Subtitr (subtitle) generatsiya qilish
IT/texnik mavzudagi nutqni tanish (dataset shu yo'nalishga moslashtirilgan)

⚠️ Cheklovlar

Model so'zlashuv uslubidagi (natural, notekis) nutqqa moslashtirilgan — diktant uslubidagi rasmiy nutqda (masalan, Common Voice uslubi) natija farq qilishi mumkin
Fon shovqini yuqori bo'lgan audio yozuvlarda aniqlik pasayishi mumkin
Juda qisqa (1 soniyadan kam) audio kliplarda ishonchli natija bermasligi mumkin
Asosan IT/texnologiya mavzusidagi so'z boyligiga moslashtirilgan — boshqa sohalarda (tibbiyot, huquq va h.k.) aniqlik pastroq bo'lishi mumkin

🏋️ Training haqida

Asosiy model: maqsudxo1ja/uz-whisper-small-stt Dataset: islomov/it_youtube_uzbek_speech_dataset — 21,000 audio klip, IT mavzusidagi YouTube videolardan (Mohir Dev)

Hyperparametrlar

Table with columns: Parametr, Qiymat
Parametr	Qiymat
Learning rate	1e-5
Train batch size	4 (effektiv: 32, gradient accumulation orqali)
Optimizer	AdamW (fused)
LR scheduler	linear, 200 warmup step
Epochs	3

Framework versiyalari

Transformers 5.13.1
PyTorch 2.13.0+cu130
Datasets 5.0.0
Tokenizers 0.22.2

📄 Litsenziya

Apache 2.0

🙏 Minnatdorchilik

Asosiy model: maqsudxo1ja/uz-whisper-small-stt
Dataset: islomov/it_youtube_uzbek_speech_dataset
Asos modeli: OpenAI Whisper

Available on FriendliAI

Dedicated Endpoints

Run this model inference on single tenant GPU with unmatched speed and reliability at scale.

Learn more

Container

Run this model inference with full control and performance in your environment.

Learn more

Model Details

Model Provider

maqsudxo1ja

Model Tree

Base

maqsudxo1ja/uz-whisper-small-stt

Fine-tuned

this model

Input Modalities

Audio

Output Modalities

Text

Supported Functionality

Dedicated EndpointsContainer

Explore FriendliAI today

Get started Talk to an engineer

README

License: apache-2.0

📊 Natijalar

Table with columns: Ko'rsatkich, Qiymat
Ko'rsatkich	Qiymat
WER (Word Error Rate)	28.64%
Validation Loss	0.2634
Model hajmi	~242M parametr (Whisper-small)

Table with columns: Training Loss, Epoch, Step, Validation Loss, WER
Training Loss	Epoch	Step	Validation Loss	WER
3.0847	0.68	400	0.3161	32.49
2.4038	1.35	800	0.2827	29.58
2.0104	2.03	1200	0.2708	30.53
1.9056	2.70	1600	0.2647	28.92
1.9695	3.00	1776	0.2634	28.64

🚀 Tezkor boshlash

O'rnatish

bash
pip install transformers torch librosa soundfile

1-usul: `pipeline` bilan (eng oson)

python
from transformers import pipeline

transcriber = pipeline(
    "automatic-speech-recognition",
    model="maqsudxo1ja/uz-whisper-small-stt-v2"
)

result = transcriber("audio_fayl.mp3")
print(result["text"])

2-usul: to'g'ridan-to'g'ri model bilan (batafsil nazorat)

python
import torch
import librosa
from transformers import WhisperProcessor, WhisperForConditionalGeneration

MODEL_ID = "maqsudxo1ja/uz-whisper-small-stt-v2"

processor = WhisperProcessor.from_pretrained(MODEL_ID, language="uzbek", task="transcribe")
model = WhisperForConditionalGeneration.from_pretrained(MODEL_ID)

device = "cuda" if torch.cuda.is_available() else "cpu"
model = model.to(device)

# Audio faylni 16kHz formatida yuklaymiz (Whisper shuni talab qiladi)
audio, sr = librosa.load("audio_fayl.mp3", sr=16000)

input_features = processor(
    audio, sampling_rate=sr, return_tensors="pt"
).input_features.to(device)

predicted_ids = model.generate(input_features, language="uzbek", task="transcribe")
transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)[0]

print(transcription)

3-usul: mikrofondan real vaqtda (namuna)

python
import sounddevice as sd
import numpy as np
from transformers import pipeline

transcriber = pipeline("automatic-speech-recognition", model="maqsudxo1ja/uz-whisper-small-stt-v2")

duration = 5  # soniya
print("🎙️  Gapiring...")
audio = sd.rec(int(duration * 16000), samplerate=16000, channels=1, dtype="float32")
sd.wait()

result = transcriber({"array": audio.flatten(), "sampling_rate": 16000})
print("Natija:", result["text"])

🎯 Qo'llanish sohalari

O'zbek tilidagi audio/video kontentni matnga aylantirish
Ovozli buyruqlar va ovozli yordamchilar
Suhbat/qo'ng'iroqlarni avtomatik transkripsiya qilish
Subtitr (subtitle) generatsiya qilish
IT/texnik mavzudagi nutqni tanish (dataset shu yo'nalishga moslashtirilgan)

⚠️ Cheklovlar

Model so'zlashuv uslubidagi (natural, notekis) nutqqa moslashtirilgan — diktant uslubidagi rasmiy nutqda (masalan, Common Voice uslubi) natija farq qilishi mumkin
Fon shovqini yuqori bo'lgan audio yozuvlarda aniqlik pasayishi mumkin
Juda qisqa (1 soniyadan kam) audio kliplarda ishonchli natija bermasligi mumkin
Asosan IT/texnologiya mavzusidagi so'z boyligiga moslashtirilgan — boshqa sohalarda (tibbiyot, huquq va h.k.) aniqlik pastroq bo'lishi mumkin

🏋️ Training haqida

Asosiy model: maqsudxo1ja/uz-whisper-small-stt Dataset: islomov/it_youtube_uzbek_speech_dataset — 21,000 audio klip, IT mavzusidagi YouTube videolardan (Mohir Dev)

Hyperparametrlar

Table with columns: Parametr, Qiymat
Parametr	Qiymat
Learning rate	1e-5
Train batch size	4 (effektiv: 32, gradient accumulation orqali)
Optimizer	AdamW (fused)
LR scheduler	linear, 200 warmup step
Epochs	3

Framework versiyalari

Transformers 5.13.1
PyTorch 2.13.0+cu130
Datasets 5.0.0
Tokenizers 0.22.2

📄 Litsenziya

Apache 2.0

🙏 Minnatdorchilik

Asosiy model: maqsudxo1ja/uz-whisper-small-stt
Dataset: islomov/it_youtube_uzbek_speech_dataset
Asos modeli: OpenAI Whisper

uz-whisper-small-stt-v2

README

📊 Natijalar

🚀 Tezkor boshlash

O'rnatish

1-usul: pipeline bilan (eng oson)

2-usul: to'g'ridan-to'g'ri model bilan (batafsil nazorat)

3-usul: mikrofondan real vaqtda (namuna)

🎯 Qo'llanish sohalari

⚠️ Cheklovlar

🏋️ Training haqida

Hyperparametrlar

Framework versiyalari

📄 Litsenziya

🙏 Minnatdorchilik

Explore FriendliAI today

README

📊 Natijalar

🚀 Tezkor boshlash

O'rnatish

1-usul: pipeline bilan (eng oson)

2-usul: to'g'ridan-to'g'ri model bilan (batafsil nazorat)

3-usul: mikrofondan real vaqtda (namuna)

🎯 Qo'llanish sohalari

⚠️ Cheklovlar

🏋️ Training haqida

Hyperparametrlar

Framework versiyalari

📄 Litsenziya

🙏 Minnatdorchilik

1-usul: `pipeline` bilan (eng oson)

1-usul: `pipeline` bilan (eng oson)