pingla_tr.py

pingla.py

from pingala_shunya import PingalaTranscriber

# Initialize with specific backend and settings
transcriber = PingalaTranscriber(
    model_name="shunyalabs/pingala-v1-en-verbatim",
    backend="ct2",
    device="cuda",
    compute_type="float16"
)

# Advanced transcription with full metadata
segments, info = transcriber.transcribe_file(
    "audio2.wav",
    language='en',
    beam_size=16,                    # Higher beam size for better accuracy
    word_timestamps=True,            # Enable word-level timestamps
    temperature=0.0,                 # Deterministic output
    log_prob_threshold=-1.0,         # Filter by probability
    no_speech_threshold=None,         # Silence detection threshold
    initial_prompt="Accurately transcribe this medical dictation",  # Guide the model
    hotwords="Excelsia",        # Boost specific words
    vad_filter=False,                 # Enable voice activity detection
    task="transcribe",                # or "translate" for translation
    condition_on_previous_text=False
)

# Print transcription info
model_info = transcriber.get_model_info()
print(f"Backend: {model_info['backend']}")
print(f"Model: {model_info['model_name']}")
print(f"Language: {info.language} (confidence: {info.language_probability:.3f})")
print(f"Duration: {info.duration:.2f} seconds")

all_text = []
for segment in segments:
    all_text.append(segment.text.strip())
# Join all parts with a space and print
full_transcription = " ".join(all_text)
print(full_transcription)

ffmpeg -i audio-sample-4.mp3 -filter:a "atempo=0.70" audio2.wav

For fast audio reduce the audio speed t0 15-30%

On Linux requires: https://github.com/Purfview/whisper-standalone-win/releases/tag/libs

Extract and set path to LD_LIBRARY_PATH env