-
Any recommendations for voice-to-text software?
I recently started using voice-to-text on my MacBook for work. Although I can still type by hand, I’m a slow typist, and my arm tires out easily these days. Now, I can work faster, more efficiently, and save my energy. Plus, it’s also great exercise for these little lungs of mine.
That said, this feature probably only understands me 70% of the time. I do have trouble annunciating some words and talk quietly, but I still try to be as articulate as I can. Even so, it’s still having trouble understanding me fully.
Are there any high-quality programs available that can translate my speech into text? Do you have any personal experience using one of them?
-
5 Replies
-
I bought this one.
It uses latest ML models and performes offline:
https://apps.apple.com/us/app/whisper-transcription/id1668083311
apps.apple.com
Quickly and easily transcribe audio files into text with state-of-the-art transcription technology Whisper. Whether you're recording a meeting, lecture, or other important audio, Whisper for Mac quickly and accurately transcribes your audio files into text. *Features - Easily record and … Continue reading
-
For Mac, the best solution is probably its own Voice Control, or Dictation (the latter has fewer features). Previously, it was very poor, but over the last couple of years, they have significantly improved it. However, for Windows, there is a much more productive solution, called Dragon Professional Individual. It is expensive (over $600) but mighty software with extensive learning capabilities, which is very important in your (and mine) case, because it should learn and adapt to your specific condition. This software also existed for Macs, but unfortunately, it was discontinued about 10 years ago. I might recommend that you install Windows on a virtual machine (Parallels Desktop or VMware Fusion – I have worked with both, and Parallels Desktop is my personal choice) and run Dragon on it. This approach has some shortcomings, such as: you will have to “type” in Windows window (no pun intended), then copy and paste it into your Mac software. Also, depending on the amount of RAM in your MacBook, it might be a little bit sluggish, but not necessarily. And, one of the most important things: what microphone do you use? For speech recognition purposes, the best solution is a headset, USB or wireless, and Dragon has certain recommendations on that subject, but sometimes their solutions are overpriced. Anyway, start by choosing and experimenting with a few microphones using the software you currently have – you might be surprised, but in many cases, it is enough to improve recognition by a factor of 10. If you have any specific questions, please don’t hesitate to drop me a line.
-
Sometime in the late aughts there were some Mac programmers who came up with MacSpeech. It had some glitches, but it was OK. I graded it a B. It was bought out by the company that makes Dragon, for PCs only. There was no Mac version of Dragon until they bought MacSpeech and changed its name to Dragon Dictate. They did not improve it. They did not fix anything with it that needed to be fixed. Then in 2018 it was discontinued. Apple announced that later that year they would release Apple Voice Control. I anticipated that. I prepared for it. When I got it, I was appalled. It was horrible. I couldn’t operate my computer with it. I had to uninstall the system upgrade and go back to Dragon just to continue my life. I waited two more years. I had started to see scuttlebutt saying that there had been much improvement. I waited a little longer before I had to take the plunge. I hate Apple Voice Control. Just as you wrote, it doesn’t understand much of what you say. Its biggest difference with Dragon/MacSpeech is that the eggheads who created it programmed it to accept one pronunciation for each of the 600,000 words in the English language. MacSpeech understands that there are many accents and dialects. When you installed MacSpeech you had to read through some stories so it could adjust itself to you with a voice profile for each individual user. Apple Voice Control doesn’t do that. In New York, khakis are pants. In Boston, that’s how they say car keys. I have written dozens of emails to the people at Apple. They don’t do anything. I grade Apple Voice Control D minus.
-
I totally agree with you about MacSpeech and will never forgive Nuance for killing the whole project. Regarding Apple voice control: People with Apple Silicon processors in their computers say that voice control has worked really smooth for them over the last couple of years. I am still on Intel Mac, and voice control is horrible for dictation, and barely useful for controlling the computer.
-
-
I feel you on this challenge. I lost my ability to type by hand about a decade ago and had to transition completely to speech recognition software.
My situation was unique. As an experienced computer programmer, I ended up taking matters into my own hands (figuratively speaking). The existing commercial solutions couldn’t understand my speech well enough to be practical:
Dragon was unusable for me despite its reputation. I use a PC, and Windows built-in speech recognition was the only one that worked at all for me, and only barely. Even after following the full training procedure (reading long passages so it could adapt to my voice), I could only make it work by spelling letter-by-letter using the NATO phonetic alphabet. Attempting actual dictation remained unusable.
But that basic functionality was enough to bootstrap something better. While using Windows speech recognition to spell everything out, I kept detailed recordings of my voice plus the exact commands I was using. This built up a corpus of my personal speech data for training. At the same time, I was adapting and combining open-source software with academic speech recognition libraries, training a custom model exclusively on my own data, essentially making me the archetypal user it was designed to understand.
The result: I’ve been using this custom system successfully for about 6 years now. It understands me absurdly well, frequently correctly recognizing utterances that sound unintelligible even in my own head after I say them.
Unfortunately, although I have packaged up and released a lot of my work as open-source software, it still requires a fairly technical background to set up and use. It’s more targeted to your average computer programmer. I haven’t yet had the time to make it more user-friendly or accessible for non-technical users. I’ve also never had someone else with SMA try my speech model, so I don’t know how well it would work, although I think there is kind of an “SMA voice”.
Links to a video demo (unfortunately I didn’t include normal dictation, even though it works very well) and the software (only includes a generic speech model not my own) are below.
Log in to reply.