Speech to Text Converter
Convert speech to text online with VexaScribe. AI-powered voice recognition supports 99 languages with automatic speaker detection. Upload any recording and get accurate, timestamped transcripts in minutes.
Supported formats:
Speech to Text (STT) vs Text to Speech (TTS) — are you in the right place?
- ✅ Speech to Text (this page): you have audio or your voice → get written text.
- ➜ Text to Speech: you have written text → get spoken audio. See a TTS tool like NaturalReaders or ElevenLabs Voice — different product entirely.
Quick test: do you have audio (STT, this page) or text (TTS, other tools)?
Speech to text converts spoken words to written text. Use browser-based dictation (click mic, speak — free, instant, Web Speech API) or upload an audio file (MP3/WAV/M4A up to 5 GB — VexaScribe uses Whisper large-v3, ~92-95% accuracy, 30 min free trial). Native alternatives: Windows Voice Access (Win+H), macOS Dictation (Fn Fn), iOS/Android system dictation.
Verified August 2026.
What is Speech to Text?
Speech to text — also called speech recognition, voice-to-text, or automatic speech recognition (ASR) — is the process of converting spoken language into written text using AI. It's used across industries to transcribe meetings, lectures, interviews, podcasts, and any audio where you need a text record.
Modern speech-to-text technology uses deep learning models trained on millions of hours of speech data. These models can handle diverse accents, speaking styles, background noise, and technical vocabulary with remarkable accuracy.
VexaScribe uses state-of-the-art AI models for speech recognition. Learn more about the technology behind our Whisper transcription and OpenAI transcription pages.
Speech recognition software vs speech-to-text: what's the difference?
Speech recognition is the underlying technology — the ML models that turn audio waveforms into words. When Whisper, wav2vec 2.0, or NVIDIA Parakeet output text from audio, that's speech recognition. The models are what academic papers benchmark.
Speech-to-text is the user-facing tool built on top: an app, a website, or a browser feature that takes your voice and gives you readable, editable text. VexaScribe, Dragon NaturallySpeaking, Otter, Google Voice Typing, and Windows Voice Access are all speech-to-text tools. They almost all run on top of one of the recognition models above.
Practical implication: when comparing accuracy, ask about the recognition model (Whisper large-v3? Nova-3? Custom?) — not the marketing brand. Different speech-to-text tools running the same model give you the same accuracy on the same audio.
Speech to text in 99 languages — accuracy by language
Whisper large-v3 word error rates on the FLEURS benchmark (lower = more accurate). VexaScribe uses this model, so these numbers apply to file upload here.
| Language | Whisper WER (FLEURS) | Practical read |
|---|---|---|
| English (US) | ~4% | Baseline — best-supported |
| Spanish | ~4-5% | Excellent, near English parity |
| French | ~5-6% | Excellent |
| German | ~5-7% | Excellent; dialects (Bairisch, Schwäbisch) drop to ~10% |
| Portuguese (BR/PT) | ~5-6% | Excellent both variants |
| Italian | ~5-6% | Excellent |
| Japanese | ~5-8% | Excellent for cleanly spoken audio |
| Chinese (Mandarin) | ~9-13% | Good; drops on Cantonese/dialects |
| Korean | ~8-11% | Good |
| Arabic (MSA) | ~10-14% | Good for MSA; regional Arabic drops significantly |
| Hindi | ~10-13% | Good |
| Indonesian | ~4% | Surprisingly excellent — top of the 99-language list |
Source: Whisper large-v3 model card + FLEURS benchmark (Google Research, 2023). Real-world accuracy varies with recording quality, accent, and background noise. All 99 supported languages are listed in the OpenAI Whisper documentation.
Speech to Text Use Cases
Meetings & Conferences
Transcribe team meetings, board calls, and conference sessions with speaker labels
Lectures & Education
Convert lectures and seminars into searchable study notes for students
Podcasts & Media
Turn podcast episodes into show notes, blog posts, and social media content
Interviews & Research
Transcribe research interviews and journalistic conversations accurately
Legal & Medical
Generate transcripts for depositions, consultations, and patient notes
Accessibility
Create captions and transcripts for deaf and hard-of-hearing audiences
How Speech to Text Technology Works
Audio Input
Your audio file is loaded and preprocessed — noise is filtered and the signal is normalized for optimal recognition.
Feature Extraction
The AI converts audio waveforms into spectrograms and extracts acoustic features that represent speech patterns.
Language Model Processing
Transformer-based neural networks match acoustic patterns to words and apply language context to improve accuracy.
Text Output
The final transcript is generated with timestamps, speaker labels, and punctuation — ready for export in your chosen format.
Sample Transcript
Works with Popular Recording Platforms
Affordable Pricing
Same rate for all audio formats and sources. No premium for speaker detection.
View pricing plans →Free Online Tools vs Professional Speech to Text
Free Browser Tools
- ✗Limited to short recordings
- ✗Basic accuracy only
- ✗No speaker detection
- ✗No file export options
- ✗Privacy concerns
Best for: Quick notes and short dictation
VexaScribe
- ✓Files up to 5GB supported
- ✓Professional AI accuracy
- ✓Automatic speaker labels
- ✓Export TXT/DOCX/SRT/VTT/JSON
- ✓Secure encrypted processing
Best for: Professional transcription of any recording
How Speech to Text Works with VexaScribe
Upload Your Recording
Drag and drop or browse to select your audio or video file. We support MP3, WAV, M4A, FLAC, MP4, and more.
AI Converts Speech to Text
Our speech recognition engine processes your audio, identifying words, speakers, and language — generating a complete timestamped transcript.
Download Your Transcript
Review and edit in our built-in editor. Export as TXT, DOCX, SRT, VTT, or JSON with all speaker labels and timestamps.
Why Choose VexaScribe for Speech to Text?
Professional speech recognition powered by the latest AI technology
State-of-the-Art Accuracy
Our AI models are trained on diverse speech data — accents, speaking speeds, technical vocabulary, and real-world audio conditions.
Fast Transcription
A 1-hour recording takes about 5-10 minutes. Upload your file and the transcript is ready before your coffee gets cold.
Speaker Diarization
Automatically identify and label different speakers. Essential for meetings, interviews, and any multi-person conversation.
99 Languages
Speech to text in 99 languages. Auto-detection identifies the spoken language, or specify it manually for optimal results.
Flexible Export Options
Download transcripts as TXT, DOCX, SRT, VTT, or JSON. Every format includes timestamps and speaker information.
Enterprise-Grade Security
All recordings are encrypted during upload and processing. Delete your files anytime. We never access or share your content.
Native OS Speech to Text — Free, No Signup
Your Windows, Mac, iPhone, or Android already has built-in speech to text. For live dictation into a text field, these are usually the fastest free option. Verified July 2026.
| OS | Tool | How to enable | Accuracy | Works in |
|---|---|---|---|---|
| Windows 11 | Voice Access / Dictation | Win + H | ~85-90% | Any text field |
| macOS | Dictation | Fn Fn or System Settings | ~88-92% | Any text field |
| iOS | Keyboard Dictation | Mic button on keyboard | ~90-93% | Any text field |
| Android | Google Voice Typing | Mic on Gboard | ~90-93% | Any text field |
When to use VexaScribe instead: you have an audio file to upload (podcast episode, meeting recording, voice memo), you need speaker labels for interviews, you need a language not covered by your OS, or you need DOCX/SRT/VTT export. Native dictation is text-field-only and single-speaker.
For developers — APIs and open-source
Building speech-to-text into your own app? You have two paths: use a cloud API or run a model locally. Both are legitimate in 2026 — the pick depends on your latency budget, privacy requirements, and volume.
Cloud speech-to-text APIs
| API | Model | Streaming | Free tier | Best for |
|---|---|---|---|---|
| OpenAI Whisper API | Whisper large-v3 | Batch only | Pay-as-you-go | Multilingual batch transcription |
| Google Cloud STT | Conformer + LLM decoder | Yes | 60 min/month | Production streaming, enterprise |
| AWS Transcribe | Amazon proprietary | Yes | 60 min/month for 12 months | AWS-integrated pipelines |
| AssemblyAI | Universal-2 | Yes | Trial credit | Speaker labels + summarization |
| Deepgram | Nova | Yes (fast) | $200 credit | Ultra-low-latency streaming |
Open-source models (run locally)
- Whisper (OpenAI) — MIT license, 99 languages, encoder-decoder. See /whisper-install-guide for setup.
- faster-whisper — Whisper reimplementation on CTranslate2. Up to 4× faster inference, int8 quantization.
- whisper.cpp — C/C++ port, runs on CPU, mobile-friendly.
- wav2vec 2.0 — Meta, self-supervised, HuggingFace Model Hub.
- NVIDIA Parakeet — RNN-T, efficient real-time on Apple Silicon via MLX.
- Kaldi — hybrid DNN/HMM toolkit, dominant in academic research.
For accuracy benchmarks across engines, see our Whisper accuracy analysis.
Looking for something more specific?
Speech to text covers a lot of ground. If your job is more specific than "convert speech to text," jump to the right page:
Turn iOS voice memos (M4A) into searchable text with speaker labels.
Drag an MP3/WAV/M4A/FLAC/OGG file and get a transcript with speaker labels.
Diarization is a separate model — the "who is speaking" task on top of ASR.
Zoom/Teams/Meet recordings — meeting bot, speaker labels, action items.
Run OpenAI Whisper on your own machine — 100% free, offline, unlimited.
Real-world WER data by language and audio condition — vendor claims vs measured.
Speech to Text FAQ
How do I turn on speech-to-text?
For live browser dictation: open Chrome, Edge, or Safari, and click the microphone icon on tools like VexaScribe or Google Docs voice typing (Chrome only, Ctrl+Shift+S). For system-wide dictation into any app: Windows uses Voice Access (Win+H), macOS uses Dictation (Fn+Fn), iOS uses the mic key on the on-screen keyboard, Android uses Gboard's mic icon. For file transcription: upload the file to VexaScribe (30 min free) or a similar tool — dictation features can't accept file input.
How do I do Google Voice to text?
Two different Google features: (1) Google Docs voice typing — open a Google Doc in Chrome desktop, Tools → Voice typing, or Ctrl+Shift+S / Cmd+Shift+S. (2) Google Cloud Speech-to-Text — the developer API used by Google Assistant and third-party apps, priced per minute. On mobile Google Docs, use the mic key on the on-screen keyboard (that's iOS or Android system dictation, not Docs voice typing).
What is the best free speech-to-text app?
For live browser dictation: Web Speech API (built into Chrome, Edge, Safari), Speechnotes, VexaScribe. For system dictation on your OS: macOS Dictation (Fn+Fn, offline on Apple Silicon), Windows Voice Access (Win+H). For file transcription: VexaScribe gives 30 minutes free; for unlimited free, install OpenAI Whisper locally (open-source, requires Python setup). On Pixel devices, Google Recorder is free with on-device processing. "Best" depends on use case — casual dictation (native OS), file transcription (VexaScribe or Whisper CLI), multilingual (Whisper).
How can I convert audio to text?
Two paths: (1) Live microphone input — click a mic button on a browser tool (VexaScribe, Speechnotes) or use OS dictation (Fn+Fn on Mac, Win+H on Windows). Text appears as you speak. (2) Existing audio file (MP3, WAV, M4A, MP4) — drag it into a file-upload transcription tool like VexaScribe (30 min free, up to 5 GB per file, 99 languages, speaker labels for up to 50 voices). Live dictation can't accept file input; file transcription can't process live mic. Different jobs, different tools.
Is speech to text free?
Yes. For live browser dictation (click mic, speak), Web Speech API is built into Chrome, Edge, and Safari — no signup, no cost. For uploaded audio files, VexaScribe gives 30 minutes free (no credit card), then $2/mo for 200 minutes. For unlimited free file transcription, install OpenAI's Whisper locally — 100% free forever, but requires Python setup. Windows Voice Access and macOS Dictation are also free built-in options for text-field dictation.
What's the difference between speech to text and text to speech?
Opposite directions. Speech to text (STT) converts your voice or a recording INTO written text — that's what this page is about. Text to speech (TTS) converts written text INTO spoken audio — that's tools like NaturalReaders, ElevenLabs Voice, or Amazon Polly. Quick test: do you have audio and want text (STT, right page)? Or do you have text and want audio (TTS, wrong page)?
Which is more accurate: browser dictation or file upload?
File upload with a Whisper-based tool (like VexaScribe) is generally more accurate — Whisper large-v3 reaches ~92-95% accuracy on clean English audio. Browser Web Speech API dictation typically runs 85-92% depending on your browser and mic. Browser wins on speed and privacy (some browsers process locally); file upload wins on accuracy, speaker labels, and file-format flexibility.
What languages are supported?
Live browser dictation (Web Speech API) supports roughly 30 languages depending on browser — Chrome and Edge have the broadest coverage. File upload with VexaScribe supports 99 languages using Whisper large-v3, with automatic language detection. Windows Dictation supports ~15 languages, macOS Dictation ~30 languages.
How accurate is speech to text?
On clean audio in supported languages: Whisper large-v3 (file upload) reaches ~92-95% word accuracy; Web Speech API (browser) typically 85-92%; native OS dictation (Windows Voice Access, macOS Dictation) 85-92%. Accuracy drops on noisy audio, heavy accents, technical jargon, or overlapping speakers. For professional transcription needs (legal, medical, publishing), human verification adds a 99%+ pass.
Can I use speech to text without downloading an app?
Yes. All the main options work in-browser: VexaScribe (upload or dictate), Speechnotes, SpeechTexter, Google Docs voice typing. No downloads, no installs. If you want offline unlimited free speech-to-text with no cloud involvement, install OpenAI's Whisper locally — that requires Python but processes everything on your computer.
What audio file formats can I upload?
VexaScribe accepts MP3, WAV, M4A, FLAC, OGG, AAC, WMA, AIFF, plus video formats (MP4, MOV, AVI, MKV, WebM) with audio auto-extracted. Maximum file size 5 GB. For very long audio, split into segments for easier review.
Does speech to text work offline?
Depends on the tool. Windows Voice Access (Win+H), macOS Dictation, and iOS/Android system dictation work offline on most modern devices with the language pack downloaded. Browser Web Speech API usually requires internet (some browsers cache language packs locally). VexaScribe file upload is cloud-based, so needs internet. Whisper installed locally works 100% offline once installed.
Is my voice recording private?
Depends on the tool. VexaScribe: files encrypted in transit and at rest, deletable at any time, never used to train AI. Browser Web Speech API: some browsers process locally (Chrome uses cloud services by default; Safari on macOS/iOS processes on-device for supported languages). Windows/Mac native dictation: on-device processing for most modern OS versions. For genuinely sensitive audio, install Whisper locally — the audio never leaves your computer.
How is this different from Windows Dictation or Mac Dictation?
Native OS dictation (Windows Voice Access, macOS Dictation, iOS/Android system dictation) is free, on-device, and works into any text field on your computer/phone — but only for LIVE dictation, no file upload. For transcribing an existing audio file (podcast episode, meeting recording, voice memo), you need a tool like VexaScribe that accepts uploads. For live dictation into a document you're writing, native OS options are usually the simplest choice.
Note: Speech to text accuracy depends on audio quality, background noise, speaker clarity, and accents. For best results, use clear recordings with minimal background noise.
VexaScribe converts speech to text from any source — meetings, lectures, interviews, podcasts, and more. Upload any audio or video file to get started.