#cmusphinx — Public Fediverse posts
Live and recent posts from across the Fediverse tagged #cmusphinx, aggregated by home.social.
-
So, for a focused voice control solution where the set of reasonable phrases is small, CMU Sphinx (pocketsphinx specifically) does not suck at all! I put together a small JSGF grammar for the kinds of things you'd say to control a podcast engine (play podcast, pause podcast, skip forward 10 minutes), and recognition is basically 100%.
Next step is to think through how to make the current grammar conditional so we can do conversations.
-
So, for a focused voice control solution where the set of reasonable phrases is small, CMU Sphinx (pocketsphinx specifically) does not suck at all! I put together a small JSGF grammar for the kinds of things you'd say to control a podcast engine (play podcast, pause podcast, skip forward 10 minutes), and recognition is basically 100%.
Next step is to think through how to make the current grammar conditional so we can do conversations.
-
So, for a focused voice control solution where the set of reasonable phrases is small, CMU Sphinx (pocketsphinx specifically) does not suck at all! I put together a small JSGF grammar for the kinds of things you'd say to control a podcast engine (play podcast, pause podcast, skip forward 10 minutes), and recognition is basically 100%.
Next step is to think through how to make the current grammar conditional so we can do conversations.
-
So, for a focused voice control solution where the set of reasonable phrases is small, CMU Sphinx (pocketsphinx specifically) does not suck at all! I put together a small JSGF grammar for the kinds of things you'd say to control a podcast engine (play podcast, pause podcast, skip forward 10 minutes), and recognition is basically 100%.
Next step is to think through how to make the current grammar conditional so we can do conversations.
-
So, for a focused voice control solution where the set of reasonable phrases is small, CMU Sphinx (pocketsphinx specifically) does not suck at all! I put together a small JSGF grammar for the kinds of things you'd say to control a podcast engine (play podcast, pause podcast, skip forward 10 minutes), and recognition is basically 100%.
Next step is to think through how to make the current grammar conditional so we can do conversations.
-
CMU Sphinx first impressions, pros cons and other:
Pros:
- Works
- Is local
- Easy integration to Python via PocketSphinx
Cons:
- Currently clocking in at "I helped Apple wreck a nice beach" levels of accuracy
- ... but I think I'm holding it wrong, and there are tuning parameters I can apply?
Other:
- Can't say the name without thinking of Venture Bros.
-
CMU Sphinx first impressions, pros cons and other:
Pros:
- Works
- Is local
- Easy integration to Python via PocketSphinx
Cons:
- Currently clocking in at "I helped Apple wreck a nice beach" levels of accuracy
- ... but I think I'm holding it wrong, and there are tuning parameters I can apply?
Other:
- Can't say the name without thinking of Venture Bros.
-
CMU Sphinx first impressions, pros cons and other:
Pros:
- Works
- Is local
- Easy integration to Python via PocketSphinx
Cons:
- Currently clocking in at "I helped Apple wreck a nice beach" levels of accuracy
- ... but I think I'm holding it wrong, and there are tuning parameters I can apply?
Other:
- Can't say the name without thinking of Venture Bros.
-
CMU Sphinx first impressions, pros cons and other:
Pros:
- Works
- Is local
- Easy integration to Python via PocketSphinx
Cons:
- Currently clocking in at "I helped Apple wreck a nice beach" levels of accuracy
- ... but I think I'm holding it wrong, and there are tuning parameters I can apply?
Other:
- Can't say the name without thinking of Venture Bros.
-
CMU Sphinx first impressions, pros cons and other:
Pros:
- Works
- Is local
- Easy integration to Python via PocketSphinx
Cons:
- Currently clocking in at "I helped Apple wreck a nice beach" levels of accuracy
- ... but I think I'm holding it wrong, and there are tuning parameters I can apply?
Other:
- Can't say the name without thinking of Venture Bros.
-
Today's step in experimenting with text-to-speech: speech-to-text!
Started with a PocketSphinx integration. It's a good start, but definitely has issues with fidelity; right now I'm getting "Newton handwriting recognition" levels of accuracy for a simple "I think you said 'x'" feedback loop query. I'll have to look into tuning it.
There's another tool I found useful that has an emacs integration and runs on the graphics card; I'll have to look into that one as well. Will share its name when I remember it.
-
Today's step in experimenting with text-to-speech: speech-to-text!
Started with a PocketSphinx integration. It's a good start, but definitely has issues with fidelity; right now I'm getting "Newton handwriting recognition" levels of accuracy for a simple "I think you said 'x'" feedback loop query. I'll have to look into tuning it.
There's another tool I found useful that has an emacs integration and runs on the graphics card; I'll have to look into that one as well. Will share its name when I remember it.
-
Today's step in experimenting with text-to-speech: speech-to-text!
Started with a PocketSphinx integration. It's a good start, but definitely has issues with fidelity; right now I'm getting "Newton handwriting recognition" levels of accuracy for a simple "I think you said 'x'" feedback loop query. I'll have to look into tuning it.
There's another tool I found useful that has an emacs integration and runs on the graphics card; I'll have to look into that one as well. Will share its name when I remember it.
-
Today's step in experimenting with text-to-speech: speech-to-text!
Started with a PocketSphinx integration. It's a good start, but definitely has issues with fidelity; right now I'm getting "Newton handwriting recognition" levels of accuracy for a simple "I think you said 'x'" feedback loop query. I'll have to look into tuning it.
There's another tool I found useful that has an emacs integration and runs on the graphics card; I'll have to look into that one as well. Will share its name when I remember it.
-
Today's step in experimenting with text-to-speech: speech-to-text!
Started with a PocketSphinx integration. It's a good start, but definitely has issues with fidelity; right now I'm getting "Newton handwriting recognition" levels of accuracy for a simple "I think you said 'x'" feedback loop query. I'll have to look into tuning it.
There's another tool I found useful that has an emacs integration and runs on the graphics card; I'll have to look into that one as well. Will share its name when I remember it.