Speech-to-Text and Transcription Models Compared

52 speech-to-text models from 17 providers, callable with one Speechall API key. File transcription rates run from $0.0121 to $1.20 per audio hour. Compare pricing, languages, streaming, speaker diarization, timestamps and subtitle formats, then open a model for its ID and integration limits. Compare models on your own audio before choosing one.

Rates below show USD credit usage per audio hour. Credit purchase fees are separate and appear in your billing quote. Batch and streaming rates can differ; check the streaming rate in the catalog before estimating live-audio costs.

Speech-to-text models by use case

Shortlists generated from the catalog rates and features below. Rates are credit usage, not accuracy rankings; test your own audio before choosing. Jump to the full catalog.

Lowest-cost file transcription

Models with the lowest published credit rate for recorded audio files.

  1. Gemini 2.5 Flash-Lite $0.0121 per audio hour
  2. Cloudflare Whisper $0.027 per audio hour
  3. Cloudflare Whisper Tiny (EN) $0.027 per audio hour
  4. Cloudflare Whisper Large v3 Turbo $0.0306 per audio hour
  5. Groq Whisper Large v3 Turbo $0.04 per audio hour

Real-time streaming transcription

Models that accept live audio over the WebSocket endpoint, ranked by streaming rate.

  1. AssemblyAI Universal-3.5 Pro $0.21 per streamed audio hour
  2. Deepgram Nova $0.348 per streamed audio hour
  3. Deepgram Nova 2 $0.348 per streamed audio hour
  4. AssemblyAI Universal 2 $0.37 per streamed audio hour
  5. Deepgram Nova 3 $0.462 per streamed audio hour

Speaker diarization for meetings and calls

Models that separate speakers in a recording, ranked by file rate.

  1. Grok Speech to Text $0.10 per audio hour
  2. Microsoft MAI-Transcribe-2 $0.10 per audio hour
  3. Soniox Async v5 $0.10 per audio hour
  4. AssemblyAI Universal 2 $0.15 per audio hour
  5. Azure AI Speech-to-Text $0.18 per audio hour

Subtitles with word timestamps

Models that return SRT and VTT subtitles and word-level timestamps, ranked by file rate.

  1. Microsoft MAI-Transcribe-2 $0.10 per audio hour
  2. Soniox Async v5 $0.10 per audio hour
  3. AssemblyAI Universal 2 $0.15 per audio hour
  4. Azure AI Speech-to-Text $0.18 per audio hour
  5. Rev AI Enhanced $0.30 per audio hour

Widest language coverage

Models with the most supported languages listed in the catalog.

  1. Azure AI Speech-to-Text 134 languages, automatic language detection
  2. Gladia Solaria 114 languages, automatic language detection
  3. ElevenLabs Scribe 81 languages, automatic language detection
  4. Microsoft MAI-Transcribe-2 61 languages, automatic language detection
  5. Soniox Async v5 60 languages, automatic language detection

Open-source models as a hosted API

Whisper and NVIDIA Parakeet are open-weight models. These providers run them, so there are no GPUs to manage.

  1. Cloudflare Whisper $0.027 per audio hour
  2. Cloudflare Whisper Tiny (EN) $0.027 per audio hour
  3. Cloudflare Whisper Large v3 Turbo $0.0306 per audio hour
  4. Groq Whisper Large v3 Turbo $0.04 per audio hour
  5. Groq Whisper Large v3 $0.111 per audio hour

Full speech-to-text model catalog

Feature filters match any selected feature. Check each model’s badges when you need several features together. Prices below estimate credit usage per audio hour; credit purchase fees are quoted separately.

Subtitle Formats Features Details
AssemblyAI Universal 2
Universal-2 is a state-of-the-art model built on Universal-1, offering enhanced ...
v2
AssemblyAI
Batch: $0.15000/hr
Streaming: $0.37000/hr
SRT VTT
Punctuation Diarization Streaming Speaker Labels Word Timestamps Language Detection Confidence
AssemblyAI Universal-3.5 Pro
Universal-3.5 Pro is AssemblyAI’s advanced speech-to-text model for complex, rea...
v3
AssemblyAI
Batch: $0.21000/hr
Streaming: $0.21000/hr
None
Punctuation Diarization Streaming Speaker Labels Word Timestamps Language Detection Confidence
Azure AI Speech-to-Text
Azure's default, general-purpose speech-to-text model, trained on a vast amount ...
Azure
Batch: $0.18000/hr
SRT VTT
Punctuation Diarization Speaker Labels Word Timestamps Language Detection Confidence
Cloudflare Whisper
A general-purpose speech recognition model based on OpenAI's Whisper, trained on...
Cloudflare
Batch: $0.02700/hr
VTT
Punctuation Language Detection
Cloudflare Whisper Large v3 Turbo
Whisper is a pre-trained model for automatic speech recognition (ASR) and speech...
Cloudflare
Batch: $0.03060/hr
VTT
Punctuation Word Timestamps Language Detection
Cloudflare Whisper Tiny (EN)
This is the English-only version of the Whisper Tiny model which was trained on ...
Cloudflare
Batch: $0.02700/hr
VTT
Punctuation Word Timestamps
Deepgram Base
Standard base model for speech recognition
v2024-01-26.8851
deepgram
Batch: $0.75000/hr
Streaming: $0.87000/hr
VTT
Punctuation Diarization Streaming Word Timestamps Language Detection
Deepgram Enhanced
Improved accuracy model for speech recognition
deepgram
Batch: $0.87000/hr
Streaming: $0.99000/hr
VTT
Punctuation Diarization Streaming Word Timestamps Language Detection
Deepgram Enhanced Finance
Enhanced model optimized for finance terminology
deepgram
Batch: $0.87000/hr
Streaming: $0.99000/hr
VTT
Punctuation Diarization Streaming Word Timestamps Language Detection
Deepgram Enhanced General
Enhanced model for general-purpose transcription
deepgram
Batch: $0.87000/hr
Streaming: $0.99000/hr
VTT
Punctuation Diarization Streaming Word Timestamps Language Detection
Deepgram Enhanced Meeting
Enhanced model optimized for meetings and conferences
deepgram
Batch: $0.87000/hr
Streaming: $0.99000/hr
VTT
Punctuation Diarization Streaming Word Timestamps Language Detection
Deepgram Enhanced Phonecall
Enhanced model optimized for phone conversations
deepgram
Batch: $0.87000/hr
Streaming: $0.99000/hr
VTT
Punctuation Diarization Streaming Word Timestamps Language Detection
Deepgram Nova
Advanced, high-performance speech recognition model
deepgram
Batch: $0.25800/hr
Streaming: $0.34800/hr
VTT
Punctuation Diarization Streaming Word Timestamps Language Detection
Deepgram Nova 2
High-accuracy, next-generation speech recognition model
deepgram
Batch: $0.25800/hr
Streaming: $0.34800/hr
VTT
Punctuation Diarization Streaming Speaker Labels Word Timestamps Language Detection
Deepgram Nova 2 Automotive
Nova 2 model optimized for automotive industry
deepgram
Batch: $0.25800/hr
Streaming: $0.34800/hr
VTT
Punctuation Diarization Streaming Speaker Labels Word Timestamps Language Detection
Deepgram Nova 2 ConversationalAI
Nova 2 model optimized for conversational AI
deepgram
Batch: $0.25800/hr
Streaming: $0.34800/hr
VTT
Punctuation Diarization Streaming Speaker Labels Word Timestamps Language Detection
Deepgram Nova 2 Drivethru
Nova 2 model optimized for drive-through scenarios
deepgram
Batch: $0.25800/hr
Streaming: $0.34800/hr
VTT
Punctuation Diarization Streaming Speaker Labels Word Timestamps Language Detection
Deepgram Nova 2 Finance
Nova 2 model optimized for finance terminology
deepgram
Batch: $0.25800/hr
Streaming: $0.34800/hr
VTT
Punctuation Diarization Streaming Speaker Labels Word Timestamps Language Detection
Deepgram Nova 2 General
Nova 2 model for general-purpose transcription
deepgram
Batch: $0.25800/hr
Streaming: $0.34800/hr
VTT
Punctuation Diarization Streaming Speaker Labels Word Timestamps Language Detection
Deepgram Nova 2 Medical
Nova 2 model optimized for medical terminology
deepgram
Batch: $0.25800/hr
Streaming: $0.34800/hr
VTT
Punctuation Diarization Streaming Speaker Labels Word Timestamps Language Detection
Deepgram Nova 2 Meeting
Nova 2 model optimized for meetings and conferences
deepgram
Batch: $0.25800/hr
Streaming: $0.34800/hr
VTT
Punctuation Diarization Streaming Speaker Labels Word Timestamps Language Detection
Deepgram Nova 2 Video
Nova 2 model optimized for video content
deepgram
Batch: $0.25800/hr
Streaming: $0.34800/hr
VTT
Punctuation Diarization Streaming Speaker Labels Word Timestamps Language Detection
Deepgram Nova 2 Voicemail
Nova 2 model optimized for voicemail transcription
deepgram
Batch: $0.25800/hr
Streaming: $0.34800/hr
VTT
Punctuation Diarization Streaming Speaker Labels Word Timestamps Language Detection
Deepgram Nova 3
Great accuracy in a broader range of real-world enterprise use cases and challen...
deepgram
Batch: $0.25800/hr
Streaming: $0.46200/hr
VTT
Punctuation Diarization Streaming Speaker Labels Word Timestamps Language Detection
Deepgram Nova 3 General
Nova 3 model for general-purpose transcription
deepgram
Batch: $0.31200/hr
Streaming: $0.55200/hr
VTT
Punctuation Diarization Streaming Speaker Labels Word Timestamps Language Detection
Deepgram Nova 3 Medical
Nova 3 model optimized for medical terminology
deepgram
Batch: $0.31200/hr
Streaming: $0.55200/hr
VTT
Punctuation Diarization Streaming Speaker Labels Word Timestamps Language Detection
Deepgram Nova General
Nova model for general-purpose transcription
deepgram
Batch: $0.25800/hr
Streaming: $0.34800/hr
VTT
Punctuation Diarization Streaming Word Timestamps Language Detection
Deepgram Nova Phonecall
Nova model optimized for phone conversations
deepgram
Batch: $0.25800/hr
Streaming: $0.34800/hr
VTT
Punctuation Diarization Streaming Word Timestamps Language Detection
ElevenLabs Scribe
Scribe is a speech-to-text model built for accuracy and handling real-world audi...
v1
ElevenLabs
Batch: $0.40000/hr
SRT VTT
Punctuation Diarization Speaker Labels Word Timestamps Language Detection
Gemini 2.5 Flash
Best model for price-performance, ideal for high-throughput tasks like large-sca...
v2.5
Gemini
Batch: $0.12222/hr
None
Punctuation Language Detection
Gemini 2.5 Flash-Lite
Most cost-efficient and fastest model, optimized for high-volume, latency-sensit...
v2.5
Gemini
Batch: $0.01215/hr
None
Punctuation Language Detection
Gemini 2.5 Pro
Most advanced model for complex tasks, excelling at coding and complex prompts.
v2.5
Gemini
Batch: $0.26100/hr
None
Punctuation Language Detection
Gemini 3.5 Transcribe
File transcription through the Interactions API with automatic language identifi...
Gemini
Batch: $0.30600/hr
None
Punctuation Diarization Speaker Labels Word Timestamps Language Detection
Gladia Solaria
Gladia's cutting-edge, next-generation ASR model, launched in April 2025. Design...
v1
Gladia
Batch: $0.61200/hr
VTT
Punctuation Diarization Speaker Labels Language Detection
Grok Speech to Text
Grok Speech to Text (STT) is xAI's standalone speech recognition API for batch a...
SpaceXAI
Batch: $0.10000/hr
None
Punctuation Diarization Speaker Labels Word Timestamps Language Detection Voice Activity Detection
Groq Whisper Large v3
A multilingual ASR model offering high accuracy and speed for transcription and ...
vv3
groq
Batch: $0.11100/hr
None
Punctuation Word Timestamps Language Detection
Groq Whisper Large v3 Turbo
A pruned and fine-tuned version of Whisper Large v3, designed for faster and les...
vv3 Turbo
groq
Batch: $0.04000/hr
None
Punctuation Word Timestamps Language Detection
IBM Watson Speech to Text
A cloud-based speech recognition service from IBM Watson that converts audio int...
IBM
Batch: $1.20000/hr
None
Punctuation Diarization Speaker Labels Word Timestamps Confidence
Microsoft MAI-Transcribe-2
Microsoft AI MAI-Transcribe-2 for multilingual batch speech-to-text using Azure ...
v2
Azure
Batch: $0.10000/hr
SRT VTT
Punctuation Diarization Speaker Labels Word Timestamps Language Detection Confidence
Mistral Voxtral Mini Transcribe
Voxtral is Mistral’s audio model family designed for powerful speech understandi...
Mistral
Batch: $0.06000/hr
None
Punctuation Language Detection
Mistral Voxtral Mini Transcribe v2
Voxtral is Mistral’s audio model family designed for powerful speech understandi...
Mistral
Batch: $0.17999/hr
None
Punctuation Language Detection
NVIDIA Parakeet TDT 0.6B V3 Realtime
NVIDIA Parakeet TDT 0.6B V3 Realtime on Together AI is a 600-million-parameter m...
v3
togetherai
Batch: $0.21000/hr
None
Punctuation Word Timestamps Language Detection
OpenAI GPT-Transcribe
File transcription with detected languages, prompt and keyword hints. No timesta...
OpenAI
Batch: $0.27000/hr
None
Punctuation Language Detection
OpenAI Whisper
General-purpose speech recognition model. Based on the open-source Whisper large...
vlarge-v2
OpenAI
Batch: $0.36000/hr
SRT VTT
Punctuation Word Timestamps Language Detection
Rev AI Enhanced
Rev AI's high-accuracy general-purpose speech-to-text model, trained on a divers...
v2.0
RevAI
Batch: $0.30000/hr
SRT VTT
Punctuation Diarization Speaker Labels Word Timestamps Language Detection Confidence
Rev AI Reverb ASR
Rev AI's open-source derived English Automatic Speech Recognition (ASR) model. T...
v1.0
RevAI
Batch: $0.30000/hr
SRT VTT
Punctuation Diarization Speaker Labels Word Timestamps Language Detection Confidence
Smallest AI Pulse
Smallest AI Pulse is a multilingual speech-to-text model for both real-time stre...
Smallest AI
Batch: $0.18000/hr
None
Punctuation Diarization Speaker Labels Word Timestamps Language Detection Voice Activity Detection Confidence
Smallest AI Pulse Pro
Smallest AI Pulse Pro is an English-only batch speech-to-text model optimized fo...
v4.1
Smallest AI
Batch: $0.24000/hr
None
Punctuation Word Timestamps Confidence
Soniox Async v5
Multilingual file transcription with diarization, word timestamps and custom con...
soniox
Batch: $0.10000/hr
SRT VTT
Punctuation Diarization Speaker Labels Word Timestamps Language Detection Confidence
Speechmatics Enhanced
Speechmatics' Enhanced ASR model offers very good accuracy, though processing is...
speechmatics
Batch: $0.75000/hr
SRT
Punctuation Diarization Speaker Labels Word Timestamps Language Detection Confidence
Speechmatics Standard
Speechmatics' Standard ASR model offers faster results with good accuracy.
speechmatics
Batch: $0.45000/hr
SRT
Punctuation Diarization Speaker Labels Word Timestamps Language Detection Confidence
Thinking Machines Inkling
Thinking Machines Lab's open-weights multimodal reasoning model hosted on Togeth...
togetherai
Batch: $0.61200/hr
None
Punctuation

How to choose a speech-to-text API model

  1. Choose your workflow: recorded files or live WebSocket transcription. Batch support does not guarantee the same features or languages in streaming.
  2. Filter for required outputs: diarization for separating speakers, word timestamps for alignment, or SRT/VTT for captions. Support varies by model.
  3. Check the model’s file-size, duration and encoding limits. Pick an audio upload method that fits your application.
  4. Test the same representative recordings with each shortlisted model. Compare transcription errors, speaker separation, latency and credit usage. There is no universal accuracy winner for every accent or recording.

How do I estimate transcription cost?

Multiply audio hours by the applicable model rate. For example, 10 hours at $0.30 per audio hour consumes $3.00 of USD credit value. This is an illustration, not a model quote. Check streaming rates separately and review the fee and currency conversion when buying credits.

Can I switch between providers?

Use a different provider-qualified model ID with the same Speechall transcription endpoint. Recheck optional parameters, languages and output formats when switching: a unified API does not make every model’s capabilities identical. See the API quickstart and OpenAI-compatible upload guide.

How does Speechall compare with OpenRouter?

Both platforms offer access to multiple transcription models. The existing Speechall and OpenRouter speech-to-text comparison explains Speechall’s interfaces and offers a checklist for live WebSocket transcription, raw audio uploads, subtitle output and reusable text replacement rules, with links to both platforms’ documentation.

What happens to my audio and transcript?

The Speechall privacy policy describes its no-storage policy for audio and transcripts after processing. Your selected provider processes the audio under its own terms. Review both before sending sensitive recordings; contact Speechall support for requirements that need confirmation.

Speech-to-text model comparison FAQ

Which speech-to-text models support real-time streaming?

Look for the Streaming badge and use the Streaming feature filter to shortlist models. Speechall provides a WebSocket endpoint at wss://api.speechall.com/v1/transcribe. Check the model details for supported streaming languages and the catalog for streaming pricing; batch capabilities do not guarantee the same behavior in a live session.

How do I choose a model for speaker diarization?

Use the Diarization and Speaker Labels badges to find candidates for meetings, interviews, and calls. Diarization separates speech by speaker; it does not establish a person’s identity. Enable diarization on a supported model and evaluate speaker separation on representative audio, especially overlapping speech.

Can I get word timestamps and SRT or VTT subtitles?

Word Timestamps, SRT, and VTT are separate capabilities in this catalog. Choose a model with each capability your application needs. For supported file transcription models, request word-level timestamp granularity with JSON output, or request output_format=srt or output_format=vtt for subtitle text. Subtitle support alone does not imply word-level timing.

How much does speech-to-text transcription cost per hour?

The catalog converts each model’s per-second credit rate to an hourly estimate by multiplying it by 3,600. For example, 0.0001 USD per second equals 0.36 USD per audio hour. Streaming rates can differ from batch rates. Credit purchase service fees are shown in the billing quote, so include those fees when estimating your total spend.

How should I compare speech recognition accuracy and latency?

Test the same representative recordings across shortlisted models and measure transcript errors, speaker separation, processing time, and cost. For live captions, also measure time to partial and final results. Language, accents, noise, recording quality, and overlapping speakers can change results; catalog feature badges are not an accuracy benchmark.

Integration and pricing guides

From model comparison to your first transcript

  1. Create a Speechall account and create an API key in the console.
  2. Review your credit balance and purchase quote. Usage draws down prepaid credits; checkout shows purchase fees and conversion.
  3. Test models with your own audio in the playground, then follow the API quickstart.