With the continuous advancement of AI, it is really easy to convert speech to text. I recently discovered this super powerful Whisper JAX tool, which supports direct microphone self-recording, uploading recording files, and YouTube URLs. I tested an 8-minute video and completed it in less than 5 seconds, and a 12-minute video in less than 10 seconds. It is super fast. The model used is the Whisper API, which is very accurate and can mix Chinese and English. It also provides the function of adding a timestamp. The key point is that it is completely free.
If you need to create subtitle files for speeches, meeting recordings, or videos in the future, you can directly throw them into this tool to quickly obtain the text content.
Whisper JAX is a powerful free speech-to-text AI tool that converts an 8-minute video in less than 5 seconds.
Whisper JAX is placed in the Hugging Face platform and shared by sanchit-gandhi user. Although the interface is only in English, the operation is very simple. Go to the homepage and select the mode you want to use, “Microphone”, “Audio File”, “YouTube”:
You can click to upload the sound file, or you can drag it directly into the window:
I tested YouTube first. After putting the YouTube URL, click Submit below and it will start processing. In addition, although it has a translate option, it seems to be unavailable at the moment, so it can only convert speech to text:
Then let it run for a while. There will be a reference number for seconds, but I don’t think it’s accurate. Most of the time it’s much faster than it predicts. For example, this YouTube video of Ada is over 8 minutes long. It says it takes 51 seconds, but it doesn’t actually take that long:
After the conversion is completed, the text content will be displayed in the transcription below. The great thing is that not only Chinese and English are OK, but also Japanese. The Japanese mentioned in Ada’s video can be recognized, but there is an error. There is a small one-click copy icon in the upper right corner, allowing you to quickly copy all content:
This video actually only took 4.8 seconds to complete, much faster than its estimated 51 seconds, so that number of seconds is just a reference, don’t be scared when you see a very long number of seconds:
Next, I tested MKBHD’s latest “Driving The New Fastest Car Ever” video, which was also transcribed quickly:
It only takes 9.7 seconds to transcribe a video of more than 12 minutes:
The transcribed Chinese text basically has no symbols in the middle, so you may find it difficult to segment it. In this case, you can turn on the “Return timestamps” function, then press Submit to transcribe, and the obtained text content will have timestamps:
According to the developer, the model used by Whisper JAX is Whisper large-v2 launched by OpenAI. I tested many videos and the accuracy reached 99.9%. Only a very small number of voices may be recognized incorrectly if they are not spoken clearly.
Source: KOCPC Chinese