Hi, I’m Seniru Epasinghe 👋

I’m a AI undergraduate and an AI enthusiast, working on machine learning projects and open-source contributions.
I enjoy exploring AI pipelines, natural language processing, and building tools that make development easier.

---

🌐 Connect with me

Sinscribe

This model is a fine-tuned version of openai/whisper-small on the Sinhala CSV + FLACs dataset. It achieves the following results on the evaluation set:

• Loss: 0.0829
• Wer: 29.1387

Intended uses & limitations

Can be used for Sinhala speech to text conversions. Make sure to input noise low audio to the model, to get the best outcome.

Training and evaluation data

Trained on the custom dataset - seniruk/sinscribe-sinhala-stt

Trained on above final dataset with 2 epochs on a device with below spec for 41:00:59 hours

• 16GB RAM
• NVIDIA GeForce RTX 4060 Ti 16GB
• 12th Gen Intel(R) Core(TM) i5-12400F (2.50 GHz) - 64-bit

Training results

Inferencing

import torchaudio
from transformers import pipeline
import torch

device = 0 if torch.cuda.is_available() else -1

from transformers import pipeline

pipe = pipeline("automatic-speech-recognition", model="seniruk/whisper-small-si",device=device)

def transcribe(audio_path):
    if audio_path is None:
        return "No audio received. Please record something."

    waveform, sample_rate = torchaudio.load(audio_path)

    if waveform.shape[0] > 1:
        waveform = waveform.mean(dim=0, keepdim=True)

    array = waveform.squeeze().numpy()

    result = pipe({"array": array, "sampling_rate": sample_rate})
    return result["text"]

result= transcribe("audio.wav")
print(result)

Gradio UI

import torchaudio
from transformers import pipeline
import gradio as gr
import torch

# === Setup ===
device = 0 if torch.cuda.is_available() else -1
MAX_DURATION_SECONDS = 30  # Limit audio length to 30 seconds

# Load fine-tuned Whisper pipeline
pipe = pipeline(
    "automatic-speech-recognition",
    model="seniruk/whisper-small-si",
    device=device
)

def transcribe(audio_path):
    try:
        if audio_path is None:
            return "No audio received. Please record or upload a file."

        # Load and prepare audio
        waveform, sample_rate = torchaudio.load(audio_path)

        # Convert to mono
        if waveform.shape[0] > 1:
            waveform = waveform.mean(dim=0, keepdim=True)

        # Duration check
        duration = waveform.shape[1] / sample_rate
        if duration > MAX_DURATION_SECONDS:
            return f"Audio too long ({duration:.1f}s). Please use a clip shorter than {MAX_DURATION_SECONDS}s."

        # Convert to numpy array
        array = waveform.squeeze().numpy()

        # Run inference
        result = pipe({"array": array, "sampling_rate": sample_rate})

        return result.get("text", "No transcription returned.")

    except Exception as e:
        return f"Error during transcription: {e}"


# === Gradio Interface ===
iface = gr.Interface(
    fn=transcribe,
    inputs=gr.Audio(sources=["microphone", "upload"], type="filepath", label="Record or Upload Audio"),
    outputs=gr.Textbox(label="Transcription"),
    title="Whisper Small Sinhala (GPU/CPU)",
    description=(
        "Sinhala speech-to-text demo using a fine-tuned Whisper Small model(Sinscribe) "
        "Supports microphone recording or file upload (max 30 seconds)"
    ),
)

iface.launch()

CPU runnable version of above model is below

• seniruk/whisper-small-si-cpu

Framework versions

• Transformers 4.48.0
• Pytorch 2.5.1+cu121
• Datasets 3.6.0
• Tokenizers 0.21.4