OpenAI Whisper API 2026: Speech-to-Text for AI Applications

Whisper is OpenAI's state-of-the-art speech recognition model, available via API and locally.

Why Whisper?

99 languages supported with high accuracy

Handles accents, background noise, and technical vocabulary

Outputs timestamps for word and segment alignment

Available via API or run locally for privacy

API Transcription

python
from openai import OpenAI
import os
client = OpenAI()
Basic transcription
with open('audio.mp3', 'rb') as f:
    transcript = client.audio.transcriptions.create(
        model='whisper-1',
        file=f,
        language='en',  # Optional - auto-detects if not specified
        response_format='text'  # text, json, srt, vtt, or verbose_json
    )
print(transcript)
Verbose JSON with timestamps
with open('meeting.mp3', 'rb') as f:
    transcript = client.audio.transcriptions.create(
        model='whisper-1',
        file=f,
        response_format='verbose_json',
        timestamp_granularities=['word', 'segment']
    )
print(f'Duration: {transcript.duration}s')
for seg in transcript.segments:
    print(f'[{seg.start:.1f}s - {seg.end:.1f}s] {seg.text}')
Word-level timestamps
for word in transcript.words:
    print(f'{word.word}: {word.start:.2f}s - {word.end:.2f}s')

Translation (Non-English to English)

python
with open('french_interview.mp3', 'rb') as f:
    translation = client.audio.translations.create(
        model='whisper-1',
        file=f,
        response_format='text'
    )
print(translation)  # Always returns English

Local Whisper (Free, Private)

bash
pip install openai-whisper
Or faster-whisper for 4x speed
pip install faster-whisper

python
faster-whisper (recommended for local use)
from faster_whisper import WhisperModel
Models: tiny, base, small, medium, large-v3
model = WhisperModel('medium', device='cuda', compute_type='float16')
CPU: model = WhisperModel('base', device='cpu', compute_type='int8')
segments, info = model.transcribe('audio.mp3', beam_size=5)
print(f'Detected language: {info.language} ({info.language_probability:.0%})')
for segment in segments:
    print(f'[{segment.start:.1f}s -> {segment.end:.1f}s] {segment.text}')

Real-Time Transcription

python
import pyaudio
import wave
import tempfile
import threading
CHUNK = 1024
FORMAT = pyaudio.paFloat32
CHANNELS = 1
RATE = 16000
RECORD_SECONDS = 5
def record_and_transcribe():
    audio = pyaudio.PyAudio()
    stream = audio.open(format=FORMAT, channels=CHANNELS, rate=RATE, input=True, frames_per_buffer=CHUNK)
    print('Recording...')
    frames = [stream.read(CHUNK) for _ in range(0, int(RATE / CHUNK * RECORD_SECONDS))]
    stream.close()
    audio.terminate()
    
    with tempfile.NamedTemporaryFile(suffix='.wav', delete=False) as f:
        wf = wave.open(f.name, 'wb')
        wf.setnchannels(CHANNELS)
        wf.setsampwidth(audio.get_sample_size(FORMAT))
        wf.setframerate(RATE)
        wf.writeframes(b''.join(frames))
        wf.close()
        
        with open(f.name, 'rb') as audio_file:
            result = client.audio.transcriptions.create(model='whisper-1', file=audio_file)
        return result.text
print(record_and_transcribe())

Meeting Transcription + AI Summary Pipeline

python
def transcribe_and_summarize(audio_path: str) -> dict:
    # Transcribe
    with open(audio_path, 'rb') as f:
        transcript = client.audio.transcriptions.create(
            model='whisper-1', file=f, response_format='verbose_json'
        )
    
    text = transcript.text
    
    # Summarize with GPT-4
    summary = client.chat.completions.create(
        model='gpt-4o',
        messages=[{
            'role': 'user',
            'content': f'Summarize this meeting transcript. Include:\n'
                       f'1. Key decisions made\n'
                       f'2. Action items with owners\n'
                       f'3. Next steps\n\nTranscript:\n{text}'
        }]
    )
    
    return {
        'transcript': text,
        'duration': transcript.duration,
        'summary': summary.choices[0].message.content
    }

Conclusion

Whisper is the most reliable speech-to-text solution in 2026. Use the API for convenience, faster-whisper locally for privacy and cost. The meeting transcription + AI summary pipeline is immediately production-ready.