Is there any speech-to-text software usable in a Fedora Linux environment?
Much thanks,
Max Pyziur pyz@brama.com
On Wed, 2023-02-22 at 18:15 -0500, Max Pyziur wrote:
Is there any speech-to-text software usable in a Fedora Linux environment?
Yes. More than one, espeak & festival, to name just two that spring to mind. Though I've only toyed with them yonks ago, so others will be of more help than me.
You should say what you're needs are. Screenreading for the vision impaired, reading out prepared text (e.g. ebooks), etc.
Max Pyziur wrote:
Is there any speech-to-text software usable in a Fedora Linux environment?
Tim:
Yes. More than one, espeak & festival, to name just two that spring to mind. Though I've only toyed with them yonks ago, so others will be of more help than me.
Oops, sorry, that was the other way around (text to speech). There is a project to do something like Alexa:
https://fedoramagazine.org/your-personal-voice-assistant-on-fedora-linux/
I don't know about dictation, though (there's more than one purpose for speech recognition).
On Thu Feb23'23 05:01:31PM, Community Support for Fedora Users wrote:
From: Tim via users users@lists.fedoraproject.org Date: Thu, 23 Feb 2023 17:01:31 +1030 To: users@lists.fedoraproject.org Cc: Tim ignored_mailbox@yahoo.com.au Reply-To: Community support for Fedora users users@lists.fedoraproject.org Subject: Re: Speech to text software in a Fedora Linux environment
On Wed, 2023-02-22 at 18:15 -0500, Max Pyziur wrote:
Is there any speech-to-text software usable in a Fedora Linux environment?
Yes. More than one, espeak & festival, to name just two that spring to mind. Though I've only toyed with them yonks ago, so others will be of more help than me.
I have never used espeak, but I have used festival a bit in the days that I used sylpheed. I used it, when I remembered to do so, to read out longer non-technical emails. It was fairly good, understandably stumbling on non-English words, but was slower than my reading.
I wonder if either of them can be used with mutt (which I use now). Also for reading articles on browsers (never done that) I notice that espeak has a very small installed RPM, compared to festival.
Ranjan
Oops, my mistake. I got carried away and did not realize that the OP was wanting speech-to-text software, even though it is in his Subject line!
Ranjan
On Thu Feb23'23 07:15:43AM, Ranjan Maitra wrote:
From: Ranjan Maitra mlmaitra@gmx.com Date: Thu, 23 Feb 2023 07:15:43 -0600 To: Community support for Fedora users users@lists.fedoraproject.org Reply-To: Community support for Fedora users users@lists.fedoraproject.org Subject: Re: Speech to text software in a Fedora Linux environment
On Thu Feb23'23 05:01:31PM, Community Support for Fedora Users wrote:
From: Tim via users users@lists.fedoraproject.org Date: Thu, 23 Feb 2023 17:01:31 +1030 To: users@lists.fedoraproject.org Cc: Tim ignored_mailbox@yahoo.com.au Reply-To: Community support for Fedora users users@lists.fedoraproject.org Subject: Re: Speech to text software in a Fedora Linux environment
On Wed, 2023-02-22 at 18:15 -0500, Max Pyziur wrote:
Is there any speech-to-text software usable in a Fedora Linux environment?
Yes. More than one, espeak & festival, to name just two that spring to mind. Though I've only toyed with them yonks ago, so others will be of more help than me.
I have never used espeak, but I have used festival a bit in the days that I used sylpheed. I used it, when I remembered to do so, to read out longer non-technical emails. It was fairly good, understandably stumbling on non-English words, but was slower than my reading.
I wonder if either of them can be used with mutt (which I use now). Also for reading articles on browsers (never done that) I notice that espeak has a very small installed RPM, compared to festival.
Ranjan _______________________________________________ users mailing list -- users@lists.fedoraproject.org To unsubscribe send an email to users-leave@lists.fedoraproject.org Fedora Code of Conduct: https://docs.fedoraproject.org/en-US/project/code-of-conduct/ List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines List Archives: https://lists.fedoraproject.org/archives/list/users@lists.fedoraproject.org Do not reply to spam, report it: https://pagure.io/fedora-infrastructure/new_issue
On Wed, Feb 22, 2023 at 6:16 PM Max Pyziur pyz@brama.com wrote:
Is there any speech-to-text software usable in a Fedora Linux environment?
I've used CMU's Sphinx suite on Ubuntu and Fedora. I've used PocketSphinx on RaspberryPi's.
My experience with Sphinx is, a lot of Type II errors. That is, it does great recognizing voice commands from a small grammar, like "dial 911". But it would recognize background noise/background television as voice command, too. It's a bad outcome when one of the actions is "dial 911".
You might want to try Mozilla's DeepSpeech. It depends on TensorFlow, and that gave me trouble in the past.
My next test of voice-to-text will be with Kaldi ASR. It's on my TODO list.
Jeff
On Thu, 2023-02-23 at 04:07 -0500, Jeffrey Walton wrote:
My experience with Sphinx is, a lot of Type II errors. That is, it does great recognizing voice commands from a small grammar, like "dial 911". But it would recognize background noise/background television as voice command, too. It's a bad outcome when one of the actions is "dial 911".
There's a lot of complaints about Alexa doing that. Unless it's able to distinguish your voice from someone else's, that problems not going to go away. There's plenty of YouTube watchers complaining about whenever some YouTube personality says "Alexa, switch off the light," their own Alexa does it.
I don't have any of those spying devices, but when I try speech control on my phone, it's very poor at recognising what I say. I notice Samsung's one is much better than Google's, but I could only use Samsung's to do something like dictate a SMS. I can easily just use a messenger app that just lets me send a recorded sound bite. It's also a right pain to work out what command it expects you to use to do what action. I prefer actually using real light switches, anyway.
On Thu, 23 Feb 2023, Jeffrey Walton wrote:
On Wed, Feb 22, 2023 at 6:16 PM Max Pyziur pyz@brama.com wrote:
Is there any speech-to-text software usable in a Fedora Linux environment?
I've used CMU's Sphinx suite on Ubuntu and Fedora. I've used PocketSphinx on RaspberryPi's.
My experience with Sphinx is, a lot of Type II errors. That is, it does great recognizing voice commands from a small grammar, like "dial 911". But it would recognize background noise/background television as voice command, too. It's a bad outcome when one of the actions is "dial 911".
You might want to try Mozilla's DeepSpeech. It depends on TensorFlow, and that gave me trouble in the past.
My next test of voice-to-text will be with Kaldi ASR. It's on my TODO list.
Much thanks for your reply and suggestions. The goal is take an audio file of, say, a presentation, and transcribe it.
Jeff
Thank you again,
Max pyz@brama.com
On 2/23/23 08:20, Max Pyziur wrote:
On Thu, 23 Feb 2023, Jeffrey Walton wrote:
On Wed, Feb 22, 2023 at 6:16 PM Max Pyziur pyz@brama.com wrote:
Is there any speech-to-text software usable in a Fedora Linux environment?
I've used CMU's Sphinx suite on Ubuntu and Fedora. I've used PocketSphinx on RaspberryPi's.
My experience with Sphinx is, a lot of Type II errors. That is, it does great recognizing voice commands from a small grammar, like "dial 911". But it would recognize background noise/background television as voice command, too. It's a bad outcome when one of the actions is "dial 911".
You might want to try Mozilla's DeepSpeech. It depends on TensorFlow, and that gave me trouble in the past.
My next test of voice-to-text will be with Kaldi ASR. It's on my TODO list.
Much thanks for your reply and suggestions. The goal is take an audio file of, say, a presentation, and transcribe it.
My goal is to someday be in a live interview, and casually work the phrase, "Alexa, unlock the front door," into the conversation.
On Thu, 2023-02-23 at 14:24 -0600, Robert Nichols wrote:
My goal is to someday be in a live interview, and casually work the phrase, "Alexa, unlock the front door," into the conversation.
More disturbing to do the opposite... ;-)
On Wed, 22 Feb 2023 18:15:42 -0500 (EST) Max Pyziur pyz@brama.com wrote:
Is there any speech-to-text software usable in a Fedora Linux environment?
I don't know if this is much use, but youtube has a speech to text function accessible by pressing c while watching a video. It isn't available for all videos and it isn't perfect, but it does well enough to get the gist of what is being said. It is really handy for speeding up presentations with the volume turned down. Turn on audio to text, then speed the video up to 1.5 - 2.0.
The reason I mention it is it might be possible to find what they are using if it is open source and get a copy.
On Wed, 2023-02-22 at 18:15 -0500, Max Pyziur wrote:
Is there any speech-to-text software usable in a Fedora Linux environment?
Much thanks,
Max Pyziur pyz@brama.com
Just last night I finished my re-installation of Fedora 37 specifically to run Whisper https://github.com/openai/whisper
Hardware: - AMD Ryzen 7 1700 with 32GB RAM - NVIDIA 1080 with 8GB
Software - Fedora 37 installation with XFCE - rpmFusion - packages: ffmpeg, conda, python3.9, and nvidia X drivers (xorg-x11- drv-nvidia-cuda.x86_64 xorg-x11-drv-nvidia-power)
Steps to configure - Had to use a graphical target, even without a monitor attached to the computer - created a conda environment with python 3.9 - pytorch (https://pytorch.org/get-started/locally/) with cuda 11.7 - whisper setup: pip install -U openai-whisper
I am able to transcribe an hour program in about 30 minutes
I'll help where I can, but no promises. I've installed this only once.
ed
On Thu Feb23'23 02:22:19PM, Ed K wrote:
From: Ed K ed@atl.org Date: Thu, 23 Feb 2023 14:22:19 -0500 To: users@lists.fedoraproject.org Reply-To: Community support for Fedora users users@lists.fedoraproject.org Subject: Re: Speech to text software in a Fedora Linux environment
On Wed, 2023-02-22 at 18:15 -0500, Max Pyziur wrote:
Is there any speech-to-text software usable in a Fedora Linux environment?
Much thanks,
Max Pyziur pyz@brama.com
Just last night I finished my re-installation of Fedora 37 specifically to run Whisper https://github.com/openai/whisper
Hardware:
- AMD Ryzen 7 1700 with 32GB RAM
- NVIDIA 1080 with 8GB
Software
- Fedora 37 installation with XFCE
- rpmFusion
- packages: ffmpeg, conda, python3.9, and nvidia X drivers (xorg-x11-
drv-nvidia-cuda.x86_64 xorg-x11-drv-nvidia-power)
Steps to configure
- Had to use a graphical target, even without a monitor attached to the
computer
- created a conda environment with python 3.9
- pytorch (https://pytorch.org/get-started/locally/) with cuda 11.7
- whisper setup: pip install -U openai-whisper
I am able to transcribe an hour program in about 30 minutes
I'll help where I can, but no promises. I've installed this only once.
ed
Just curious, why did you have to reinstall F37 specifically to run Whisper?
Ranjan
On Thu, 2023-02-23 at 22:32 -0600, Ranjan Maitra wrote:
Just curious, why did you have to reinstall F37 specifically to run Whisper?
Ranjan
Longish story: I had whisper running on my main development laptop which is an AMD Ryzen 9 5900HX. Originally I was able to install the ROCm tool-chain on this 'unsupported' chip. And whisper took 10 minutes to transcribe an hour's recording. After a kernel update, the ROCm libraries failed to run and that same hour of recording tool 23 hours to transcribe.
I must have re-installed F37 on this laptop 5 times, trying combinations of kernels and python environments. But nothing worked. So, I decided to try nVidia - which required transitioning to my desktop. It was still running F31, since I don't use it except for some back end processing.
Borrowing two nVidia 1080s from a friend, I installed F37 on the desktop and followed the instructions for whisper. And it worked, first time, without additional hassle. (I'm a fan of AMD, but they need to get their act together on their GP-GPU support). Only one of the 1080s runs during whisper, I have not found out how to get both to process.
Now it takes about 30 minutes where the laptop took 10.
ed