We are currently using Bing Speech with LUIS, but looking to convert to Speech service.

Right now we have multiple recorders that operate in the browser, Flash, WebRTC, HTML5.

Each of these has to connect to Bing Speech to Text to get realtime translation and LUIS results to drive actions in the application. Additionally we are currently streaming the audio to Amazon S3. Ideally we would like to stream the audio only once, and have it picked up by Microsoft from Speech to Text AND be able to retrieve a URL for later use.

Having to maintain two streams has led to a number of problems where one isn't a consistent recording, and the transcription we get from Bing doesn't match what we have in our recording.

We would use this recording for review in the application, and as a possible training cohort to improve Speech to Text results using Custom Dictionary and Custom Acoustic Models.

Is this possible currently?

If not, is this on Microsoft's roadmap?

This is a big enough problem for us that we are considering moving compute/AI platforms one way or the other, especially as we have to reimplement to Speech Service by end of year.

we are using https://docs.microsoft.com/en-us/azure/cognitive-services/speech-service/quickstart-js-browser
for filing user data on screen based on Voice input (recognizeOnceAsync)..but i am facing multiple problems in that
1. There is no way around to stop recognizer immediately once i speak some words ..it take 4-5 seconds to stop mic processing ..
2. assume i want to enter user first name based on voice ..but it takes 5-6 to complete the process ,which is not recommendable..need to your inputs to improve it .
3. Any results returned from Speech service is appended with extra (.) in last of result . not sure why its happening .

we are in process of completing POC for automated application Form fill application ,so need some solution asap

I am doing a POC with speech recognition for long speeches.
https://docs.microsoft.com/de-de/azure/cognitive-services/speech/concepts#recognition-modes

The recognition mode "conversation" with format "detailed" delivers message responses of type "SpeechPhrase" including confidence value.

The recognition mode "dictation" with format "detailed" delivers message responses of type "SpeechFragment" and "SpeechPhrase" (including confidence value). But the fragments do not contain any information about confidence value.
With the C# service library and the recognition mode "dictation" you'll get partial results with a confidence value (enum). But this is not our desired solution, because the confidence value seems to belong to the whole phrase (Confidence: Indicates the level of confidence of a recognized phrase., https://cdn.rawgit.com/Microsoft/Cognitive-Speech-STT-ServiceLibrary/master/docs/html/9d706b3a-8d1f-ba71-d628-fff00928c72d.htm)

The recognition mode "interactive" is not optimized for long speeches.

A confidence number value per word or per speech fragment would be very interesting for us. Because with this confidence value it would be possible to get self-assessments of the Microsoft speech service, if it recognizes the word or the fragment correctly or not. Unfortunately I didn't found a such possibility.

Thanky you for a short answer, if there are any solution supported by Microsoft speech service.