Beginner 13 minAudio

Audio-to-text transcription: free, without limit

An open laptop casts green light across a dark desk.
Direct response

For free, unlimited audio-to-text transcription, install local software based on Whisper, OpenAI's speech recognition model released under the MIT license: Buzz or Vibe (Windows, macOS, Linux). The file stays on your computer, with no account and no quota. Free online tools have caps: 30 minutes per month on Wave, 300 minutes per month in Word with a Microsoft 365 subscription (official pages as of November 18, 2026).

Online transcription services are handy for a short file, but they impose a quota and receive your recording. Software installed on your machine transcribes as many hours as you want, in French, without sending anything. This guide helps you choose the right tool for your use case, run your first transcription, and check the result.

By Mohamed Meguedmi·Update 2026-11-18·Tested on Windows, macOS, and Linux

#How do you transcribe audio to text for free and without limits?

Truly free, unlimited audio-to-text transcription requires software installed on your computer that runs a speech recognition model locally. The only cap becomes your machine's processing time: no metered minutes, no sign-up, and no file uploads.

Automatic transcription (also called speech recognition, or speech-to-text) is the conversion of recorded speech into written text by an artificial intelligence model. The free apps covered here are built on Whisper (Vibe also offers other models). According to its official repository, Whisper is a general-purpose, multilingual speech recognition model that can also translate into English and identify the spoken language. OpenAI publishes its code and weights under the MIT license, which permits both personal and commercial use.

The Whisper repository doesn't specify any maximum file length. According to its documentation, the transcription function reads the entire file and processes it in sliding 30-second windows. So a two-hour interview is transcribed just like a two-minute memo, only it takes longer.

Which use case, which free transcription option (official pages checked November 18, 2026)
Your use caseRecommended optionWhy
A short file with no sensitive contentFree online toolNothing to install, but a monthly quota and the file is sent to the service
Interviews, meetings, confidential consultationsBuzz or Vibe locallyOffline transcription, no quota, no account
Video subtitlesBuzz (SRT, VTT) or Vibe (SRT, VTT, DOCX…)Direct export to subtitle formats
Multiple speakersBuzz or Vibe (speaker identification)Both apps advertise speaker separation
Dozens of hours, automated processingWhisper or faster-whisper on the command lineScriptable, batch processing
Live microphone dictationBuzz (real-time) or Vibe (microphone)Offline microphone transcription

#Free online audio-to-text transcription: what are the limits?

The complete kit for building your own AI assistant.

Benefit from our experience and a guided path to build your local assistant and use your AI every day.

  • Lifetime online access
  • PDF + files
  • Refunded within 30 days

Almost all free online audio transcription tools limit volume, and all of them receive your file on their servers. They're fine in a pinch for a short, non-confidential recording; beyond that, the quota or data concerns become a dealbreaker.

Wave
According to Wave's pricing page, checked on November 18, 2026, the free plan covers 30 minutes of recording per month. The unlimited Pro plan costs $15 per month or $80 per year on the web.
Microsoft Word
According to Microsoft's help page, accessed on November 18, 2026, a Microsoft 365 subscriber can transcribe up to 300 minutes of uploaded audio per month. A note on the same page says it is currently available in Word for Microsoft 365 on Windows for "Commercial Tenants," as opposed to government environments, which are limited to Word for the web; the page doesn't specify what applies to a Personal or Family subscription. Recordings are stored on OneDrive.
Google Docs
Google Docs voice typing lets you dictate into the microphone in Chrome, Edge, or Safari. Google's help page states that the browser controls the speech recognition service; the page describes dictation, not the import d of an audio file.
Other free services
Free quotas for specialized services change often, and their pricing pages block automated reading: check the maximum length per file, the number of files per day, and how long recordings are retained before you upload a document.

#Which free audio-to-text transcription software should you install?

If you don't want to use the command line, Buzz and Vibe are the most straightforward free, open-source audio transcription apps: one window, a file to drag in, text to export. Whisper, whisper.cpp, and faster-whisper suit users who are comfortable with the terminal.

Free, local transcription software, based on their official repositories as of November 18, 2026
SoftwareVersion and dateOperating systemsLicenseStrengths
Buzz1.4.5, August 23, 2026Windows, macOS (Apple Silicon and Intel), LinuxMITFiles, YouTube links, live mic, speakers, watched folder, TXT, SRT, VTT export
Vibe3.3.5, November 13, 2026Windows, macOS, LinuxMITBatch files, speakers, export to SRT, VTT, TXT, HTML, PDF, JSON, DOCX, Whisper, Nemotron 3.5, and Parakeet TDT v3 models
Whisper (OpenAI)PyPI package 20250625Windows, macOS, Linux (Python)MITOfficial reference, command line and Python
whisper.cpp1.9.5, October 6, 2026Windows, macOS, LinuxMITDependency-free C/C++, CPU-only possible, Metal on Mac, Vulkan
faster-whisperPyPI package 1.2.1Windows, macOS, Linux (Python)MITUp to 4 times faster than Whisper, according to its authors

Buzz is a desktop app that, according to its repository, transcribes and translates audio offline on your computer using Whisper. Version 1.4.5, the latest release as of November 18, 2026, is available for Apple Silicon Macs and Intel Macs; the repository warns that it is the last release for Intel Macs, as Buzz now requires an Apple Silicon chip. On Windows, the installer is not signed and Windows displays a warning; the repository says to click More info, then Run anyway (Informations complémentaires, then Exécuter quand même on a French-language Windows).

Vibe is an open-source transcription app whose repository promises fully offline transcription, with no data ever leaving the device. Vibe publishes installers for Apple Silicon Macs and Intel Macs, Windows, and Linux. Vibe also supports recording from a phone via a QR code, while the processing still runs on the computer.

#How do you transcribe audio to text with Buzz or Vibe?

Transcribing audio to text with Buzz or Vibe takes five steps: install the app from its official page, choose a model, import the file, specify the language, then export the text. Expect a few minutes of setup, plus the model download the first time.

  1. 01
    Install the official app
    Download Buzz from the SourceForge link listed in its GitHub repository (or via Flatpak and Snap on Linux), or Vibe from its official download page. Avoid mirror sites not referenced by the projects.
  2. 02
    Choose a model that fits your machine
    In the app's settings, select a Whisper model. The small model works on a laptop without a graphics card; turbo or large need more memory. The model is downloaded only once, then reused offline.
  3. 03
    Import the audio or video file
    Drag the file into the Buzz or Vibe window. Both apps accept audio and video; Vibe also accepts multiple files at once for batch processing.
  4. 04
    Set the language to French
    Choose French rather than automatic detection: since the language is known, this keeps the transcription from misdetecting it at the start of a noisy recording or one that opens with music.
  5. 05
    Export and proofread
    Export to TXT for plain text, to SRT or VTT for subtitles, or to DOCX with Vibe. Proofread proper names, numbers, and passages where several people talk at once.

On Linux, the Buzz repository provides the following Flatpak install command. For command-line Whisper, the official documentation says to install the Python package and the ffmpeg tool, then run the transcription with the turbo model.

Terminal (Linux): install Buzz
flatpak install flathub io.github.chidiwilliams.Buzz
Terminal: official Whisper (macOS with Homebrew)
pip install -U openai-whisper
brew install ffmpeg
whisper audio.flac audio.mp3 audio.wav --model turbo
→
Check that everything worked
A successful transcription produces timestamped text that follows the recording from start to finish. Listen to three random passages, at the beginning, middle, and end: if the text drifts out of sync or repeats a sentence in a loop, run it again with a larger model or by forcing the language.

#What machine do you need for local audio transcription?

Local audio transcription works on an ordinary computer, even without a graphics card: whisper.cpp supports CPU-only inference, and the small model needs less than 1 GB of memory there. A graphics card or an Apple Silicon Mac significantly cuts the wait time.

Memory required per Whisper model, according to the openai/whisper (graphics card) and whisper.cpp (RAM) repositories; relative speeds measured by OpenAI in English on an A100 card
ModelSettingsVRAM with openai/whisperMemory with whisper.cppPublished relative speed
tiny39 Mabout 1 GBabout 273 MBabout 10 times large
base74 Mabout 1 GBabout 388 MBabout 7 times large
small244 Mabout 2 GBabout 852 MBabout 4 times large
medium769 Mabout 5 GBabout 2.1 GBabout 2x large
large1 550 Mabout 10 GBabout 3.9 GBreference
turbo809 Mabout 6 GBunpublishedabout 8 times large

To estimate processing time, the faster-whisper repository publishes a benchmark on 13 minutes of audio. With the small model on an Intel Core i7-12700K CPU (8 threads), openai/whisper takes 6 min 58 s, whisper.cpp 2 min 05 s, and faster-whisper in int8 1 min 42 s. With the large-v2 model on an 8 GB NVIDIA RTX 3070 Ti card, faster-whisper in fp16 takes 1 min 03 s. These figures are published by the faster-whisper authors, using older versions of the tools (faster-whisper 1.1.0, whisper.cpp 1.7.2); your time will depend on your machine and the recording.

On Apple Silicon Macs, whisper.cpp runs on the GPU via Metal and can offload part of the work to the Neural Engine with a Core ML build. On PCs, acceleration uses CUDA for NVIDIA cards; Buzz also supports Vulkan for most other cards, including integrated ones. Don't mix the two: a CUDA setting doesn't apply to a Mac.

#Is audio-to-text transcription reliable in French?

French audio-to-text transcription is supported by Whisper's multilingual models, but its quality depends on the language, the model, and the recording. The official repository says so itself: Whisper's performance varies widely by language. French is on the model's official list of languages.

Model
According to OpenAI, the turbo model is an optimized version of large-v3 that is faster with minimal loss of accuracy. It isn't trained for translation: to translate into English, use medium or large.
Forced language
The official documentation shows the --language option with the example --language Japanese; for a recording in French, specify French, or select French in Buzz or Vibe.
Audio capture
A mic close to the speaker, with no background music, gives the model a cleaner signal. Buzz offers voice separation before transcription for noisy recordings.
Proper nouns and acronyms
Names of people, brands, and acronyms are frequent sources of errors: plan a targeted proofread, especially for meeting notes meant to be shared.
i
Claimed, published, observed
The speed and memory figures in this guide are those published by the projects themselves. No testing was carried out for this guide: only a spot-check listen on your own files will tell you whether the quality is good enough for your use.

#With local transcription software, what still goes online?

With local transcription software, the audio and text stay on your computer, but some secondary features connect to the internet. Knowing which ones lets you keep processing of sensitive files truly offline.

Model downloads
The first launch downloads the model you chose. After that, transcription no longer needs a connection.
Video site links
Buzz and Vibe can transcribe a YouTube video from its link: the app downloads the video from the internet, then transcribes it locally.
Vibe summaries
Vibe offers summaries through the Claude API, an online service that then receives the transcript text. The same software's Ollama option runs the analysis locally.
Automatic updates
Vibe advertises automatic updates, which contact the developer's server. They don't involve your files, but a fully air-gapped machine won't receive them.
Word and OneDrive
By contrast, Word's transcription stores the recording on OneDrive, and Microsoft warns that how long it takes depends on your internet connection speed: avoid it for a confidential interview.
!
Recording someone
A local tool protects the file, not the legal framework around the recording: inform the people being recorded before a meeting or interview, and store the transcript as carefully as the audio.
Frequently asked questions
Is there free audio-to-text transcription with no sign-up?+
Yes. Buzz and Vibe install and run without any account, and neither does command-line Whisper. Audio-to-text transcription then happens on your computer, free and with no quota. Online, no-sign-up services generally limit the duration or number of files you can transcribe.
Is local audio transcription really unlimited?+
Local audio transcription has no minute quota: the only limits are your machine's compute time and disk space. Whisper processes audio in 30-second windows, so a recording several hours long transcribes without manual splitting; it just takes longer.
Can you transcribe audio to text live?+
Yes. Buzz transcribes the microphone in real time and offers a presentation window to display it during an event; Vibe also transcribes the microphone and system audio. For live speech recognition, Vosk is another offline option, covered in a dedicated guide on the site.
How long does it take to transcribe an hour of audio?+
It depends on the model and the machine. The benchmark published by faster-whisper shows 1 min 42 s for 13 minutes of audio with the small model in int8 on an Intel Core i7-12700K, and 1 min 03 s with large-v2 on a RTX 3070 Ti. By simple proportion (60 ÷ 13 ≈ 4.6), an hour of audio would take roughly 8 and 5 minutes respectively on comparable machines: that's an estimate, not a published measurement.
Can you use Whisper, Buzz, or Vibe for professional work?+
Yes. Whisper's code and weights are released under the MIT license, and Buzz and Vibe are also MIT-licensed, which permits commercial use. For sensitive data, local transcription means you don't send the recording to a third-party provider, but you still need consent from the people being recorded.
Do you need a graphics card to transcribe audio to text?+
No. According to whisper.cpp, Whisper's small model runs on CPU alone with less than 1 GB of memory. A NVIDIA GPU or an Apple Silicon Mac speeds up transcription and lets you run turbo comfortably, which needs about 6 GB of VRAM with openai/whisper.

#Sources and verification as of November 18, 2026

The versions, licenses, features, memory requirements, and quotas cited were collected on November 18, 2026 from the projects' official repositories and the vendors' help and pricing pages; online service quotas and software versions may change at any time.

Did this guide help you?

Feedback, an error, or a clarification? Let us know—it improves the guide for everyone.