Trying to use "pocketsphinx_continuous" and "pocketsphinx_tidigits" speech recognition from bash commandline on a zaurus 3200 with a wm8750 audio controller.
Basically not getting much of any recognition at all. Thinking this is due to my alsa configuration.
I can record using sox fairly well if I increase the gain with a command such as:
rec -S -c1 -r24k test.wav vol 120
However, I'm not sure how to increase volume gains for "arecord" and/or "pocketsphinx_continuous".
"arecord" produces very silent garbled recordings.
I already have my input source maxed out in alsamixer. Any ideas how to set the default gain of alsa higher?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
>Decoder is more or less resistant to volume level unless its very small but in this case increase will not help you much because information is lost already.
That's good to know. Seems my driver is the issue then, and I'm in the wrong forum for this.
>About batch, pocketsphinx_test is the example on how to apply it, I suggest you to look inside.
Thanks. I'll look into it further. Hope you won't mind further questions as they arise.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
there must be a way to increase the gain that pocketsphinx receives.
Alternatively, I'm a little confused how to use a "batch" mode to recognize words from a prerecorded wave file. Could anyone help with the later if not the prior?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
> there must be a way to increase the gain that pocketsphinx receives.
Decoder is more or less resistant to volume level unless its very small but in this case increase will not help you much because information is lost already. You need to find a way to fix your driver first.
About batch, pocketsphinx_test is the example on how to apply it, I suggest you to look inside.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Trying to use "pocketsphinx_continuous" and "pocketsphinx_tidigits" speech recognition from bash commandline on a zaurus 3200 with a wm8750 audio controller.
Basically not getting much of any recognition at all. Thinking this is due to my alsa configuration.
I can record using sox fairly well if I increase the gain with a command such as:
rec -S -c1 -r24k test.wav vol 120
However, I'm not sure how to increase volume gains for "arecord" and/or "pocketsphinx_continuous".
"arecord" produces very silent garbled recordings.
I already have my input source maxed out in alsamixer. Any ideas how to set the default gain of alsa higher?
>Decoder is more or less resistant to volume level unless its very small but in this case increase will not help you much because information is lost already.
That's good to know. Seems my driver is the issue then, and I'm in the wrong forum for this.
>About batch, pocketsphinx_test is the example on how to apply it, I suggest you to look inside.
Thanks. I'll look into it further. Hope you won't mind further questions as they arise.
> Hope you won't mind further questions as they arise.
Sure, you are welcome to ask.
Well, probably the issue is in rate conversion with a broken driver. Linux sound is still hardly usable.
Can you record with arecord at 22 kHz?
Yes, I can record with arecord at 22kHz, but it's very silent and somewhat distorted.
there must be a way to increase the gain that pocketsphinx receives.
Alternatively, I'm a little confused how to use a "batch" mode to recognize words from a prerecorded wave file. Could anyone help with the later if not the prior?
> there must be a way to increase the gain that pocketsphinx receives.
Decoder is more or less resistant to volume level unless its very small but in this case increase will not help you much because information is lost already. You need to find a way to fix your driver first.
About batch, pocketsphinx_test is the example on how to apply it, I suggest you to look inside.