Speech to text explained: examples, best models and apps, ASR, accents, IPA and Tacotron

Speech to text converts spoken audio into written text: live captions, voice typing, meeting transcripts, voice assistants, call analytics and searchable recordings all use it. Modern systems are deep learning models trained on huge amounts of speech, and they now handle many languages and accents well. I built live speech-to-text into the OCR and Speech Workspace, so this page covers examples, choosing models and apps, ASR versus dictation, VS Code voice input, accents and phonetics, IPA in text to speech, Tacotron 2, and a tutorial.

What are examples of speech to text?

Everyday examples include voice typing on phone keyboards, live captions in video calls and on Android and iPhone, meeting transcription in Teams, Zoom and Google Meet, voice assistants such as Google Assistant and Siri, YouTube auto-captions, and WhatsApp voice-message transcripts.

Business examples: transcribing customer support calls for quality analysis, medical dictation, court and lecture transcription, subtitles for videos, and voice search in apps. In education, students use transcription to turn lectures into searchable notes. In my OCR and Speech Workspace, live speech-to-text sits alongside OCR so spoken and written material can both be searched and questioned.

What is the best speech-to-text model in 2026?

There is no single best model; quality depends on language, accent, audio quality, domain vocabulary and latency needs. Strong options include OpenAI's Whisper family and newer transcription models, NVIDIA's open ASR models, and commercial services from Google, Deepgram, AssemblyAI and others.

Also asked as: best speech to text model 2026

Choose by testing: record twenty to fifty clips representative of your users, including accents and background noise, transcribe them with each candidate, and compute word error rate. Public leaderboards such as Hugging Face's Open ASR Leaderboard help shortlist. For Indian languages and code-mixed Hindi-English speech, test models specifically trained on Indian data, since general models vary widely.

What is the best speech-to-text app for Android?

For most users, Google's built-in options are excellent: Gboard voice typing for dictation in any app, Live Transcribe for real-time captions of conversations, and the Recorder app on Pixel phones for transcribed recordings. Samsung phones have their own voice input and recorder transcription.

Also asked as: best speech to text app for android

Choose by use: Gboard for typing messages and notes; Live Transcribe for accessibility and following conversations; recorder apps for lectures and interviews. Download offline speech packs for private, faster recognition without internet. For long recordings with speaker labels and summaries, dedicated transcription apps add features, but check their privacy policies before uploading recordings.

What is the difference between ASR and speech to text?

There is essentially no difference: ASR, automatic speech recognition, is the technical term for the technology, and speech to text is the everyday name for what it does. Both mean converting spoken language into text automatically.

Also asked as: asr vs speech to text

Researchers and engineers say ASR when discussing models, accuracy metrics such as word error rate, and system design. Product pages say speech to text or transcription. Related terms include speaker diarisation, identifying who spoke when, and spoken language understanding, extracting meaning and intent from speech. Text to speech, TTS, is the reverse direction, generating audio from text.

What is the difference between dictation and speech to text?

Dictation is a specific use of speech to text: speaking to type text live, usually in a document or message, with voice commands for punctuation and editing, such as "comma", "new line" or "delete that". Speech to text is the broader technology that also transcribes recordings, meetings and calls.

Also asked as: dictation vs speech to text

Dictation prioritises low latency and accuracy for one speaker close to the microphone, and integrates with keyboards and editors. Transcription of recordings handles multiple speakers, longer audio and timestamps. Professional dictation software for medicine and law adds specialised vocabularies. For students, phone dictation is a quick way to capture ideas and draft assignments. Speaking a first draft and then editing it by keyboard is often faster than typing from scratch.

How do I use speech to text in VS Code?

Install Microsoft's VS Code Speech extension, which adds voice input for Copilot Chat and dictation into the editor, processing speech locally on your machine. Then use the microphone button in chat or the dictation command to speak.

Also asked as: vs code speech to text

This helps with accessibility, describing coding tasks to AI assistants by voice, and writing comments or documentation hands-free. Configure the speech language in settings. Operating system dictation features, such as Windows voice typing, also work in VS Code. Dictating code itself is awkward, but dictating instructions to an AI coding assistant works well.

Is machine learning a viable tool to map accent from speech onto text or syllables?

Yes. Machine learning models can transcribe speech into phonemes, the sound units of language, rather than ordinary spelling, capturing how something was actually pronounced, and other models classify accents or measure pronunciation differences. These are active research and product areas.

Also asked as: is machine learning a viable tool to map accent from speech onto text/syllables

Phoneme recognition models, often fine-tuned from self-supervised speech models, output IPA-like sequences that reveal accent features, useful for language learning, pronunciation scoring and speech therapy. Accuracy depends on training data covering the accents involved, and phonetic labels themselves vary between annotators. Combining phoneme recognition with forced alignment tools gives syllable-level timing.

Is there text-to-speech software that accepts IPA phonetic transcription?

Yes. Many text-to-speech systems accept IPA through SSML's phoneme tag, which specifies pronunciation using the IPA alphabet, including major cloud TTS services. Open-source engines such as eSpeak NG also support phonetic input.

Also asked as: is there a text-to-speech software that accepts ipa-based phonetic transcription

SSML phoneme tags are useful for names, technical terms and words in other languages that TTS mispronounces, such as Indian names in English speech. Support for exact IPA symbols varies by engine and voice, so test the pronunciations. Some neural TTS systems also let you add custom lexicons, mapping words to phonetic spellings once and reusing them.

How many spectrogram frames per input character does Tacotron 2 generate?

Tacotron 2 does not use a fixed ratio. Its attention mechanism learns alignments, so each character maps to a variable number of mel-spectrogram frames, typically several, depending on the sound, speaking rate and pauses.

Also asked as: how many spectrogram frames per input character does text-to-speech (tts) system tacotron-2 generate

Tacotron 2 predicts mel spectrograms with a frame hop of 12.5 milliseconds, about 80 frames per second. At a speaking rate of roughly 12 to 15 characters per second, that averages around five to seven frames per character, with vowels and pauses taking more frames than short consonants. The decoder generates frames autoregressively until a stop token predicts the end.

What is a simple speech-to-text tutorial?

A simple tutorial uses OpenAI's open-source Whisper model in Python: install it, load a model size suited to your hardware, and transcribe an audio file, getting text and timestamps.

Also asked as: tutorial speech to text

import whisper
model = whisper.load_model("small")
result = model.transcribe("lecture.mp3", language="en")
print(result["text"])
for seg in result["segments"][:3]:
    print(seg["start"], seg["end"], seg["text"])

Install with pip install openai-whisper and have ffmpeg available. Faster implementations such as faster-whisper speed up transcription on CPUs and GPUs. For live transcription, stream audio in short chunks. Choose the model size by hardware: tiny and base run on almost any laptop, while medium and large models are far more accurate but need a GPU for reasonable speed. Always check transcripts of names, numbers and technical terms, and add a short vocabulary prompt describing the topic to improve recognition of domain words.

Pranjul Rathour

GenAI Engineer · India

GenAI engineer and AI product builder with 2+ years shipping production-grade AI systems: RAG pipelines, fine-tuned LLMs, hybrid retrieval and multi-modal apps across vision, speech and OCR, architected end to end from ingestion to deployment. Leads engineering for SCULT INDIA's 14-member team, founded the 500+ member TechVerse Enclave community and has mentored 200+ students. Three hackathon first prizes: Changethon 2025 (IIT Roorkee), Product Genesis at Vividhotsava 2025 (CSJMU) and BYTEBATTLE (MeetKats).

Open to GenAI roles, hackathon judging, mentorship sessions and guest talks at colleges. For campus sessions the inviting institution covers the visit end to end: travel (a First AC train ticket for journeys up to five hours, a flight for anything longer), accommodation, meals, and pickup and drop; the honorarium and workshop fees are negotiable on WhatsApp or on the call, and requests are welcome from institutions that can cover the visit this way. Email pranjulrathour41@gmail.com.

PortfolioInvite me to your campusHire meLinkedInGitHubXInstagramThreadsFacebookBlueskyDev.toBloggerHashnodeSCULTEmail