Web Speech API not working in Edge v. 147

Hello Kelly,

It has been working poorly for over a month now. Some days it worked for a few hours. But most days the speech recognition API gives a "network" error over and over again. Below is the information you requested:

edge://metrics-internals/#variations

There is not "SpeechNetworkStack" entry.

"CFRSpeechServiceProtectECS": {

"enableFeatures": [

"msSpeechSynthesisServiceUseCetoService",

"msSpeechRecognitionServiceUseCetoService"

]

},

edge://histograms/

- Histogram: Microsoft.SpeechRecognition.CancellationReason recorded 29 samples, mean = 0.8 (flags = 0x41) [#]

0 ------O (6 = 20.7%)

1 -----------------------O (23 = 79.3%) {20.7%}

2 ...

- Histogram: Microsoft.SpeechRecognition.Error recorded 35 samples, mean = 26.3 (flags = 0x41) [#]

0 ...

25 ------------O (12 = 34.3%) {0.0%}

26 O (0 = 0.0%) {34.3%}

27 -----------------------O (23 = 65.7%) {34.3%}

28 ...

- Histogram: Microsoft.SpeechRecognition.ErrorRate recorded 149 samples, mean = 2.6 (flags = 0x41) [#]

0 ------------------------------O (30 = 20.1%)

1 ------------------------------O (30 = 20.1%) {20.1%}

2 ------------------------------O (30 = 20.1%) {40.3%}

3 ------------------------------O (30 = 20.1%) {60.4%}

4 ...

7 -----------------------------O (29 = 19.5%) {80.5%}

8 ...

- Histogram: Microsoft.SpeechRecognition.Event recorded 75 samples, mean = 1.5 (flags = 0x41) [#]

0 ------------------------------O (30 = 40.0%)

1 ------------------------------O (30 = 40.0%) {40.0%}

2 --O (2 = 2.7%) {80.0%}

3 -O (1 = 1.3%) {82.7%}

4 ------O (6 = 8.0%) {84.0%}

5 ...

8 ------O (6 = 8.0%) {92.0%}

9 O (0 = 0.0%) {100.0%}

- Histogram: Microsoft.SpeechRecognition.Method2 recorded 66 samples, mean = 1.5 (flags = 0x41) [#]

0 ------------------------------O (30 = 45.5%)

1 ------O (6 = 9.1%) {45.5%}

2 O (0 = 0.0%) {54.5%}

3 ------------------------------O (30 = 45.5%) {54.5%}

4 O

- Histogram: Microsoft.SpeechRecognition.SdkInstall recorded 30 samples, mean = 1.0 (flags = 0x41) [#]

0 O (0 = 0.0%)

1 ------------------------------O (30 = 100.0%) {0.0%}

2 ...

- Histogram: Microsoft.SpeechRecognition.WasRateLimited recorded 30 samples, mean = 0.0 (flags = 0x41) [#]

0 ------------------------------O (30 = 100.0%)

1 ...

I did some additional tests and got some observations.

I used Speech Translator extension for the tests, basically it restarts the speech recognition onend()  or onerror() events.

On the first start it returned me "No speech" error this time (so it means it started normally and there was just no sound), but on the restart I kept getting "Network error" over and over again.

Then I tried again (complete new initialization of SpeechRecognition) and I got "Network error" right away and kept getting "Network error" over and over again.

I tried a bit later and got "No speech" 2 times and then "Network Error" again.

I did some tests with different VPNs and without any - in my case it didnt affect the behavior.

And I dont have any "Microsoft.SpeechRecognition" records here:

edge://metrics-internals/#variations

The only related to SpeechRecognitions lines there are:

"CFRSpeechServiceProtectECS": { "enableFeatures": [ "msSpeechSynthesisServiceUseCetoService", "msSpeechRecognitionServiceUseCetoService" ] },

I recoded histogram for the attempt where I got 2 "No Speech" and then many "Network Error" (full report sent in dm):

Histograms The page will show the difference in histogram data since you switched to Monitoring mode. The page will automatically refresh every 1 second unless you click stop. Include histograms from subprocesses - Histogram: Microsoft.SpeechRecognition.CancellationReason recorded 2 samples, mean = 1.0 (flags = 0x41) [#] 0 O (0 = 0.0%) 1 --O (2 = 100.0%) {0.0%} 2 ... - Histogram: Microsoft.SpeechRecognition.Error recorded 4 samples, mean = 26.0 (flags = 0x41) [#] 0 ... 25 --O (2 = 50.0%) {0.0%} 26 O (0 = 0.0%) {50.0%} 27 --O (2 = 50.0%) {50.0%} 28 ... - Histogram: Microsoft.SpeechRecognition.ErrorRate recorded 20 samples, mean = 2.6 (flags = 0x41) [#] 0 ----O (4 = 20.0%) 1 ----O (4 = 20.0%) {20.0%} 2 ----O (4 = 20.0%) {40.0%} 3 ----O (4 = 20.0%) {60.0%} 4 ... 7 ----O (4 = 20.0%) {80.0%} 8 ... - Histogram: Microsoft.SpeechRecognition.Event recorded 8 samples, mean = 0.5 (flags = 0x41) [#] 0 ----O (4 = 50.0%) 1 ----O (4 = 50.0%) {50.0%} 2 ... - Histogram: Microsoft.SpeechRecognition.Method2 recorded 8 samples, mean = 1.5 (flags = 0x41) [#] 0 ----O (4 = 50.0%) 1 ... 3 ----O (4 = 50.0%) {50.0%} 4 O (0 = 0.0%) {100.0%} - Histogram: Microsoft.SpeechRecognition.SavedAudioLength recorded 4 samples, mean = 2.0 (flags = 0x41) [#] 0 ... 2 ----O (4 = 100.0%) {0.0%} 3 ... - Histogram: Microsoft.SpeechRecognition.SdkInstall recorded 4 samples, mean = 1.0 (flags = 0x41) [#] 0 O (0 = 0.0%) 1 ----O (4 = 100.0%) {0.0%} 2 ... - Histogram: Microsoft.SpeechRecognition.WasRateLimited recorded 4 samples, mean = 0.0 (flags = 0x41) [#] 0 ----O (4 = 100.0%) 1 ...

So overall it looks like some issue related to rate limiting...