Continuous Speech Recognition Android - Without Gaps

Question

I have an activity that implements RecognitionListener. To make it continuous, every time onEndOfSpeech() I start the listener again:

speech.startListening(recognizerIntent);

But, it takes some time (around half a second) till it starts, so there is this half a second gap, where nothing is listening. Therefore, I miss words that were spoken in that time difference.

On the other hand, when I use Google's Voice input, to dictate messages instead of the keyboard - this time gap does not exist. Meaning - there is a solution.

What is it?

Thanks

score 5 · Answer 1 · edited May 23 '17 at 10:30

5

I'll recommend using CMUSphinx to recognize speech continuously. To achieve continuous speech recognition using google speech recognition api, you might have to resort to a loop in a background service which will take too much resources and drains the device battery.

On the other hand, Pocketsphinx works really great. It's fast enough to spot a key phrase and recognize voice commands behind the lock screen without users touching their device. And it does all this offline. You can try the demo.

If you really want to use google's api, see this

edited May 23 '17 at 10:30

Community

1
1

answered May 17 '15 at 16:49

Nana Ghartey

7,901
1
24
26

1

Thanks, though Pocketsphinx does not work well for natural - speech dictation (it works great for a limited grammar - great for commands), and doesn't the solution in the link to google's api result in the same problem as I posted? – Ronen Rabinovici May 18 '15 at 08:31
1

How do I download pocketsphinx – Ruchir Baronia Jan 22 '16 at 01:45
@RuchirBaronia Download from the official site http://cmusphinx.sourceforge.net/ – Nana Ghartey Jan 22 '16 at 10:08
How can I implement it in a service? Thanks! – Ruchir Baronia Feb 13 '16 at 21:13
do you have example code for android because i useing this and i add hot words like ok light but it still tregir on others words – Hamza Aug 31 '17 at 02:31

score 2 · Answer 2 · answered May 17 '15 at 01:44

There are options like:

intent.putExtra(RecognizerIntent.EXTRA_SPEECH_INPUT_COMPLETE_SILENCE_LENGTH_MILLIS, 2000); // value to wait

or

intent.putExtra(RecognizerIntent.EXTRA_SPEECH_INPUT_POSSIBLY_COMPLETE_SILENCE_LENGTH_MILLIS, 2000);

These ceased to work on Jelly Bean and above, but work on ICS and below - not sure if intended or a bug!

score 2 · Accepted Answer · answered May 17 '15 at 19:07

2

try looking at a couple other api's....

speech demo : has source here and is discussed here and operated on CLI here

you could use the full duplex google api ( its rate capped at 50 per day )

Or if you like that general idea check ibm's watson discussed here

IMO - its more complex but not capped .

answered May 17 '15 at 19:07

Robert Rowntree

6,230
2
24
43

http streaming to 3rd party services seems the only solution at this point. – Ronen Rabinovici Jun 06 '15 at 19:12

Continuous Speech Recognition Android - Without Gaps

3 Answers3

Linked