PhoWhisper: Vietnamese speech recognition from VinAI

by VinAI Research

Don’t want to install PhoWhisper? Drop the file here.

The first 20 minutes of any recording are transcribed free: a recording up to 20 minutes transcribes in full, a longer one is previewed, and a pack from $6 (500 minutes) transcribes the rest and opens every export, and credits never expire.

Whisper fine-tuned on 844 hours of Vietnamese with diverse regional accents, published by VinAI Research in five sizes from 39M to 1.55B parameters.

Linux, macOS, Windows  ·  free

Get PhoWhisper Official site & pricing

TL;DR

PhoWhisper is Whisper fine-tuned for Vietnamese on 844 hours with regional accents, in five sizes (39M–1.55B), BSD-3-Clause. PhoWhisper-large reports 4.67 WER on VIVOS.

It is a model, not an app: you bring the GPU, the chunking and the subtitle export. VinAI's wider speech work is on the VinAI page.

Category
Open source
License
BSD-3-Clause
Stars
★ 259
Last push
2024-11-12
Pricing
free (BSD-3-Clause)
Platforms
Linux, macOS, Windows

What it is

PhoWhisper is a family of Vietnamese speech-recognition models from VinAI Research, presented at ICLR 2024 (Tiny Papers). The multilingual Whisper was fine-tuned on an 844-hour Vietnamese dataset covering diverse accents and released in five sizes: tiny (39M), base (74M), small (244M), medium (769M) and large (1.55B), under BSD-3-Clause. The README's word error rates for PhoWhisper-large are 8.14 on Common Voice Vietnamese, 4.67 on VIVOS, 13.75 on VLSP 2020 task 1 and 26.68 on task 2. It runs with the Hugging Face transformers pipeline.

Best for: Developers transcribing Vietnamese, including regional accents, who can run a Whisper-size model themselves.
Watch out for: Vietnamese only, a 2024 research release with no updates since November 2024, and the example pipeline expects 16 kHz audio. Hosting, long-file chunking and output formats are yours to build.

Vietnamese recordings, without hosting a model

  1. Drop the audio or video file into the box at the top of this page, or paste a link (YouTube, a podcast or RSS feed, Google Drive, Twitch or Kick). Nothing to install and no GPU needed.
  2. Vietnamese is detected automatically and the transcript comes back in Vietnamese. WhipScribe does not translate.
  3. Read the summary free. A pack from $6 (500 minutes) opens the full transcript and the downloads: TXT, DOCX, SRT, VTT and JSON. Packs are one-time purchases and credits never expire.
  4. Files can be up to 10 hours or 5 GB per file.

Related: Transcribe Vietnamese · Vietnamese transcription · VinAI speech research · SRT generator · Pricing

Install / use

pip install transformers torch  # model: vinai/PhoWhisper-small

Key specs

Sizes
39M, 74M, 244M, 769M, 1.55B
Training audio
844 h Vietnamese
WER (large, VIVOS)
4.67
Licence
BSD-3-Clause

What people actually do with PhoWhisper-style transcription

The tool is the means. These are the jobs — each one priced at published rates, each one wired up on its own page.

Transcribe a whole back catalogue in one runThe archive is the asset, but one-at-a-time upload means it never gets done.…1,000 hours ≈ $480, onceTranscribe every new file in a Drive folderFiles accumulate in a folder nobody opens, and the knowledge in them is invisible to search.…≈ $0.29 per audio hour on the $24 packCut short-form clips out of a webinar automaticallyThe clips worth posting are buried somewhere in a 60-minute recording and finding them by scrubbing costs more than the…60-min webinar ≈ $0.72

Wire it into Amazon S3n8nMake.comGoogle DriveZapierSlack

Features

Speaker diarizationNot stated
Word-level timestampsNot stated
Streaming / real-timeNot stated
Languages supportedVietnamese
HIPAA eligibleNo

Links

PhoWhisper vs Whipscribe

FeaturePhoWhisperWhipscribe
CategoryOpen sourceTranscription APIs
Pricingfree (BSD-3-Clause)$6–$24 one-time packs (credits never expire) · first 20 minutes free
Speaker identificationNoNamed in the summary, key quotes and chapters
Word timestampsNoYes
StreamingNoNo
LanguagesVietnamese99
PlatformsLinux, macOS, WindowsWeb, API, MCP

Alternatives to PhoWhisper

OpenAI Whisper
OpenAI

The reference open-source multilingual ASR model from OpenAI.

OSS · MIT ★ 108.7k
whisper.cpp
Georgi Gerganov

C/C++ port of Whisper — runs on anything, from a Raspberry Pi to Apple Silicon.

OSS · MIT ★ 53.5k
faster-whisper
SYSTRAN

4× faster than reference Whisper using CTranslate2 — production sweet spot.

OSS · MIT ★ 25.3k

Frequently asked about PhoWhisper

Which PhoWhisper size should I use?

PhoWhisper-small (244M) is a practical start on a modest GPU; PhoWhisper-large (1.55B) scored best in every benchmark in the README, for example 8.14 WER on Common Voice Vietnamese against 11.08 for small.

How do I run PhoWhisper?

With Hugging Face transformers: pipeline('automatic-speech-recognition', model='vinai/PhoWhisper-small'), passing 16 kHz audio. Long recordings need chunking.

What licence is PhoWhisper under?

BSD-3-Clause. VinAI asks you to cite the PhoWhisper paper when it is used to produce published results.

Can WhipScribe transcribe Vietnamese?

Yes. Vietnamese is detected automatically and the transcript, summary and subtitles come back in Vietnamese. WhipScribe does not run PhoWhisper, so compare a short clip if accent accuracy is critical.

What is PhoWhisper?

PhoWhisper is Whisper fine-tuned for Vietnamese on 844 hours with regional accents, in five sizes (39M–1.55B), BSD-3-Clause. PhoWhisper-large reports 4.67 WER on VIVOS.

Is PhoWhisper free?

Yes. PhoWhisper is free and open source (free (BSD-3-Clause)).

Does PhoWhisper work on Mac, Windows and Linux?

PhoWhisper runs on Linux, macOS and Windows.

How do I install PhoWhisper?

With pip: `pip install transformers torch # model: vinai/PhoWhisper-small`. You need Python and, for most audio tools, ffmpeg on your PATH first.

Is PhoWhisper HIPAA compliant?

PhoWhisper does not claim HIPAA compliance. For PHI, run it on infrastructure you control or choose a service that will sign a BAA.

What are the limitations of PhoWhisper?

Vietnamese only, a 2024 research release with no updates since November 2024, and the example pipeline expects 16 kHz audio. Hosting, long-file chunking and output formats are yours to build.

Who is PhoWhisper best for?

Developers transcribing Vietnamese, including regional accents, who can run a Whisper-size model themselves.

What should I know before choosing PhoWhisper?

It is a model, not an app: you bring the GPU, the chunking and the subtitle export. VinAI's wider speech work is on the VinAI page.

Does PhoWhisper run offline?

PhoWhisper runs on your own machine — it is open source, so nothing leaves the computer unless you configure it to.

Can I use PhoWhisper in the browser?

No — PhoWhisper is a Linux, macOS and Windows application, not a web app. You install it rather than sign in to it.

Where is the PhoWhisper source code?

On GitHub at github.com/VinAIResearch/PhoWhisper. It is maintained by VinAI Research.

Whipscribe is a hosted transcription service. If you want transcripts without running infrastructure, paste a URL or drop a file in the form below — you'll have a transcript in seconds.

Transcribe your recordings

Running PhoWhisper yourself? Send the same file here and compare the transcript. Transcribe any language, any recording · Transcription services from $0.29 an hour · Transcribe from any meeting or recording platform · How to transcribe, step by step

Explore

All transcription tools · By language · By platform · By file format · How-to guides · Transcription services · Transcription API · Pricing

Have the recording already?

Skip the setup and send the file or the link: see transcribing a long MP3 or M4A, how to transcribe a podcast and the SRT file generator.

Comparing tools before you pay?

If you would rather pay a service than run software yourself: see an honest roundup of the best transcription services, cheap transcription services and the real cost per hour and how our transcription services work.