Speech-to-Text Shortlists

Best Free APIs and Open‑Source Engines for Speech‑to‑Text

For recurring free hosted allowances, look at Google Cloud Speech-to-Text V1 Standard or Azure Speech F0 real-time recognition. Deepgram and AssemblyAI supply one-time introductory developer credits. Whisper and Vosk run locally on your own infrastructure. Speech is Cheap provides a bounded browser demo, while its API requires a paid plan.

How to Choose

Written for software developers building prototypes without initial API expenditure, as well as builders seeking to evaluate an isolated audio sample directly within a web browser.

We include two recurring hosted quotas, two introductory credit offers, two self-hosted tools, and our bounded browser demo. Each option has verified terms. We exclude strictly paid APIs lacking verified free tiers and never label self-hosted software as a free cloud service.

  • Nature of the free tier: recurring monthly allowance, one-time trial credit, local open-source code, or browser testing.
  • Usage constraints: specific eligible endpoints, billing reset cycles, concurrency caps, and required payment methods.
  • Implementation effort: setup time, local hardware obligations, and the transition path toward production use.

Options are organized by structure: recurring hosted allowances first, followed by introductory credits, self-hosted engines, and our browser demo. Placement reflects category and operational model rather than promotional score.

Recurring Free Hosted Allowances

These cloud services provide recurring monthly usage quotas at no cost. Neither option offers unlimited transcription, and both enforce strict monthly quota boundaries.

Recurring Free API Allowance

Google Cloud Speech-to-Text V1

Best For: Prototyping lightweight applications with the legacy V1 API

Google Cloud explicitly includes 60 free minutes each month in its official V1 Standard pricing schedule.

Strengths
The 60-minute monthly allowance applies to V1 Standard speech recognition whether data logging is enabled or opted out.
Limitations
This allowance applies solely to V1 Standard, not to V2 or Chirp 3 models. Usage beyond 60 minutes and auxiliary cloud services incur standard charges.
Product Boundary
Valid only for Google Cloud Speech-to-Text V1 Standard endpoints. General cloud trial credits are a separate promotional offer.
Cost and License
60 eligible minutes per month free, followed by standard V1 per-minute rates.
Compare Google API Versions and Costs

Recurring Free API Tier

Azure Speech F0

Best For: Prototyping real-time speech recognition within Microsoft Azure

Azure Speech provides a documented recurring monthly allowance with clearly designated boundaries around its free real-time tier.

Strengths
The F0 tier grants 300 minutes of real-time transcription monthly, shared between Standard and Custom recognition endpoints.
Limitations
Batch transcription is excluded from the F0 tier. The monthly allowance cannot be applied to asynchronous batch jobs or fast transcription.
Product Boundary
Applies exclusively to Azure Speech resources configured under the free F0 pricing tier. Broader Azure trial credits remain a distinct offer.
Cost and License
300 free real-time minutes monthly. Standard charges apply if you switch to paid tiers.

Primary Sources

Compare Azure Speech Products

Introductory Credits

Developer credits allow you to test paid cloud endpoints during initial development. While these balances may not expire, they represent finite allocations that deplete with usage.

One-Time Introductory Credit

Deepgram

Best For: Evaluating hosted live and recorded speech recognition without upfront commitment

The introductory balance lets you test both live streaming and recorded batch endpoints before committing budget to ongoing production usage.

Strengths
Provides $200 in initial credit upon account creation, requiring no credit card to begin testing API requests.
Limitations
The $200 balance is a one-time trial credit, not a recurring monthly allowance. API calls consume funds according to selected models and features.
Product Boundary
Valid for hosted developer accounts. Recorded file transcription and real-time audio streams deplete the credit pool at different rates.
Cost and License
$200 one-time credit, followed by standard pay-as-you-go pricing.

Primary Sources

Check Deepgram Rates and Features

One-Time Introductory Credit

AssemblyAI

Best For: Testing core transcription alongside speech understanding tools

AssemblyAI provides a transparent trial credit balance across supported speech services, simplifying initial prototyping before you add payment details.

Strengths
Direct accounts receive $50 in non-expiring credit for eligible speech services. Any unused balance remains available after upgrading to paid billing.
Limitations
The credit does not renew monthly. LLM Gateway usage is excluded, and you must register a payment method once the credit is exhausted.
Product Boundary
Terms apply to direct accounts created on AssemblyAI. Registrations via AWS Marketplace follow different trial terms and billing rules.
Cost and License
$50 introductory credit. Usable minutes vary based on chosen model and enabled add-ons.
Check AssemblyAI Rates and Features

Free Software You Run Yourself

Open-source engines have no hosted quota limits or vendor billing. However, your team remains responsible for providing server hardware, electrical power, and long-term maintenance.

Self-Hosted Open-Source Model

OpenAI Whisper

Best For: Multilingual transcription on self-managed infrastructure

Use this option when control over the machine and offline operation matter more than setup effort.

Strengths
Downloadable code and weights let you transcribe without a hosted API account.
Limitations
Model size affects hardware needs. Users must install, run, and maintain the software.
Product Boundary
Open-source Whisper is distinct from OpenAI paid API services. An MIT license provides free code, not free cloud compute.
Cost and License
Zero license fees for weights or code. You pay for underlying compute and maintenance.
Compare Self-Hosting and an API

Self-Hosted Open-Source Engine

Vosk

Best For: Offline local speech recognition running on basic CPU hardware

Choose Vosk when developing embedded or offline desktop applications that must run locally without cloud dependencies or GPU accelerators.

Strengths
Compact language models and streaming recognition allow local speech processing on standard CPUs, mobile operating systems, and single-board computers.
Limitations
Recognition behavior and licensing terms depend on the specific model downloaded. You must package, distribute, and maintain the runtime alongside your application.
Product Boundary
A local offline recognition library, not a cloud API service. Model licensing and capabilities vary by individual language download.
Cost and License
No usage fees when self-hosted. Hardware provisioning and system maintenance remain your responsibility.

A Free Browser Demo

When you only need to evaluate a quick audio recording manually, you can test transcription directly in a browser without creating developer credentials.

Browser Demo, Paid API

Speech is Cheap

Best For: Testing sample transcription without creating a vendor account

Try our web demo to inspect transcription quality, timestamps, and speaker labels before evaluating paid API tiers.

Strengths
Upload audio, record directly, or paste a direct HTTPS media URL. It returns speakers, word timings, and downloads without signup.
Limitations
Clips must be 0.1 through 3 minutes inclusive, with file sizes at most approximately 0.0098 GiB. No API key or credits are provided.
Product Boundary
A cloud-backed browser tool for manual testing. Our API is strictly paid, requiring an active payment method at checkout.
Cost and License
The browser demo is free. Production API access requires a paid subscription or PAYG plan.
Try the Free Transcription Demo

Choose What Happens When the Allowance Ends

Before launching a prototype, check how each provider handles exhausted quotas. Determine whether your application should throttle incoming requests, hold recordings in an internal queue, or fail over gracefully to a paid API tier.

For production workloads, evaluate ongoing pricing at your projected monthly recorded volume. If self-hosting fits your technical requirements, start with our open-source shortlist. If you simply need a single quick transcript, use our browser demo.

Sources and Verification

Written and Published by Speech is Cheap. Updated .

Speech is Cheap publishes this guide and sells a service discussed in it. Recommendations rely on workload criteria rather than composite scores.

Dates show when we verified each source, not original documentation publication dates. We review sources quarterly and recheck them before updates.

  1. AssemblyAI Billing and Pricing

    Checked

  2. Azure Speech Free Tier

    Checked

  3. Deepgram Pricing

    Checked

  1. Google Cloud Speech-to-Text Pricing

    Checked

  2. OpenAI Whisper Repository

    Checked

  3. Speech is Cheap Free Browser Demo

    Checked

  1. Speech is Cheap Pricing

    Checked

  2. Vosk Models and Licenses

    Checked

  3. Vosk Offline Recognition

    Checked

Choose a Plan for Your Recordings

Review included minutes, overage rates and optional features before you integrate. Pick the Speech is Cheap subscription or pay-as-you-go plan that matches your monthly recorded volume.

Editorial and Billing Notes

Stated promotional credits are listed in USD. Azure publishes five hours of real-time recognition per month, converted here to 300 minutes. Google Cloud V1 publishes its standard free allowance directly as minutes.

Our browser demo limit of 10 MiB is approximately 0.0098 GiB, and six seconds equals 0.1 minute. One GiB equals 1,073,741,824 bytes. Stated conversions are rounded for readability; original vendor terms govern.

Free tiers and introductory promotions remain subject to vendor change. No provider listed here guarantees an immutable free hosted service. Trial credits do not renew, and self-hosted software requires your own infrastructure.

We reviewed published provider documentation and public account terms. We did not establish competitor accounts or benchmark individual user allowances for this comparison.