alumae
commented
8 years ago This is expected. I think it has to do with the dithering in the MFCC extraction, which is random. In fact, I have been confused with the same problem myself, but after debugging I concluded that it's expected.
Hi
When using the client.py script repeatedly on the same wavefile, i get different results.
Most of the time the difference is in the likelihood field of the returned hypotheses, they vary a little from one run to the next (on the same decoder)
Occasionally, the likelihood differences effect the sort order, so on subsequent calls the most likely hypothesis changes.
Is this expected? thanks
run 1: RESPONSE:{u'status': 0, u'segment-start': 0.0, u'segment-length': 4.52, u'total-length': 4.52, u'result': {u'hypotheses': [ {u'likelihood': 145.337, u'transcript': u'minus one twelve point ten minus one eleven point ten'}, {u'likelihood': 143.347, u'transcript': u'hi minus one twelve point ten minus one eleven point ten'}, {u'likelihood': 140.417, u'transcript': u'minus one twelve point ten minus one eleven point eight ten'}, {u'likelihood': 138.427, u'transcript': u'hi minus one twelve point ten minus one eleven point eight ten'}], u'final': True}, u'segment': 0, u'id': u'ceaeba27-d042-4aef-b576-781fc0eeb252'}
run 2: RESPONSE: {u'status': 0, u'segment-start': 0.0, u'segment-length': 4.52, u'total-length': 4.52, u'result': {u'hypotheses': [ {u'likelihood': 143.316, u'transcript': u'hi minus one twelve point ten minus one eleven point ten'}, {u'likelihood': 140.427, u'transcript': u'minus one twelve point ten minus one eleven point ten'}, {u'likelihood': 138.364, u'transcript': u'hi minus one twelve point ten minus one eleven point eight ten'}, {u'likelihood': 135.475, u'transcript': u'minus one twelve point ten minus one eleven point eight ten'}], u'final': True}, u'segment': 0, u'id': u'8fb2a55e-0d24-4707-8f1f-37e90c038643'}