OpenAI Whisper — State-of-the-Art ASR

OpenAI Whisper

Model	Parameters	Speed	WER (EN)
tiny	39M	32× realtime	~14%
base	74M	16× realtime	~11%
small	244M	6× realtime	~9%
medium	769M	2× realtime	~7%
large-v3	1550M	1× realtime	~5%

import whisper

model = whisper.load_model("base")

result = model.transcribe(
    "interview.mp3",
    language="en",
    task="transcribe",    # or "translate" → always outputs English
    fp16=False,
    word_timestamps=True,
)

print(f"Language: {result['language']}")
print(f"Transcript:\n{result['text']}")

for segment in result["segments"]:
    print(f"[{segment['start']:.2f}s → {segment['end']:.2f}s] {segment['text']}")

from faster_whisper import WhisperModel

# 4× faster, less VRAM, same accuracy
model = WhisperModel("large-v3", device="cpu", compute_type="int8")

segments, info = model.transcribe(
    "lecture.mp3",
    beam_size=5,
    language="en",
    vad_filter=True,
    vad_parameters=dict(min_silence_duration_ms=500),
)

print(f"Language: {info.language} ({info.language_probability:.2f})")
for segment in segments:
    print(f"[{segment.start:.2f}s - {segment.end:.2f}s] {segment.text.strip()}")

TIP: Production tip: Use faster-whisper with compute_type="int8" for CPU deployment. Runs Whisper large-v3 in real-time on modern CPUs — no GPU required.