ACS Call Automation Misses Final Speech When Call Ends Immediately

Question

ACS Call Automation Misses Final Speech When Call Ends Immediately

Kino Matuguina 20

We are using Azure Communication Services (ACS) Call Automation to make outbound calls to our contact center provider (Talkdesk). Our ACS bot does not play any prompts or speak. It only listens and sends DTMF tones in response to Talkdesk's IVR.

We are encountering an issue in the final step of our test flow. When we attempt to recognize speech, and Talkdesk ends the call immediately after responding, we receive the following error:

ERROR - An error occurred: (8522) Call not found.
Code: 8522
Message: Call not found.

It seems that if Talkdesk ends the call right after providing a spoken response, the recognize action fails, and we do not receive any transcription of what was said.

Our questions:

Is there a way for ACS to still return the recognized speech if the callee ends the call shortly after speaking, assuming the audio was already received?

We are using this setup for automated IVR testing and need to confirm each step in the call flow. Any guidance or best practices would be appreciated.

Accepted answer

0 additional answers

Your answer

Answer 1

Bhargavi Naragani 6,050 Microsoft External Staff Moderator

Hi @Kino Matuguina,

It seems like you're using Azure Communication Services (ACS) Call Automation to interact with Talkdesk’s IVR system. Your ACS bot listens for speech and sends DTMF tones in response. Everything works until the final step: Talkdesk provides a voice response and then immediately ends the call.

The error typically means the ACS platform has lost track of the call because it was terminated before the speech recognition process could complete. Since ACS needs the call to remain active while it processes audio data, the instant disconnection causes it to drop the speech recognition task and return that error.

To ensure you capture the final speech from Talkdesk before the call ends, we recommend the following approach:

Make sure you're listening for the RecognizeCompleted event and handling it as soon as it's raised. This helps in processing the recognition before the call gets disconnected. If the call ends before this event is fired, you won’t get any result.

If you have control over Talkdesk’s IVR or can work with their team, request a short delay (e.g., 1–2 seconds) after speaking the final phrase before the call is dropped. This gives ACS enough time to capture and process the final audio input.

If you’re open to another strategy, you can enable call recording and post-process the final recording to analyze speech. This is especially useful for IVR testing when exact timing is unpredictable.

Gather user input
Handle events with Event Processor
Call Recording overview

I hope you find this information useful. Let me know if you have any further questions.

Kino Matuguina 20 Reputation points

2025-05-13T08:34:48.9+00:00

Hi @Bhargavi Naragani ,

Thank you for your response.

Just to clarify. isn’t this effectively a limitation on the Microsoft side? Specifically, is there any way to still retrieve the recognized speech even if the call ends immediately after the audio is received?

We are already listening for the RecognizeCompleted event, but the issue is that the call is disconnected before this event is triggered, which prevents us from capturing the recognized speech.

Is there any recommended workaround that would allow us to receive partial or final recognition results even if the call ends abruptly?

Thank you!
Bhargavi Naragani 6,050 Reputation points Microsoft External Staff Moderator

2025-05-13T08:54:28.0733333+00:00

Hi Kino Matuguina,

yes, this behavior is a limitation of the current Azure Communication Services (ACS) Call Automation platform. The RecognizeCompleted event is only triggered if the call remains active long enough for ACS to process and return the recognition result. If the callee hangs up immediately after speaking, ACS may not have enough time to Finish processing the audio, Complete the recognition action, and deliver the result via event or callback.

So, once the call ends, ACS discards in-progress recognition operations, and there's currently no built-in support to return partial or final speech results after disconnection.

If possible, configure Talkdesk to add a brief delay (1–2 seconds) after the final spoken message before disconnecting the call. This is the most effective and direct way to give ACS enough time to return the RecognizeCompleted event reliably.

If adding a delay isn’t feasible, a more robust solution is to Enable call recording using ACS, download the recorded audio of the call and use Azure Speech-to-Text to transcribe it offline. This provides full control and ensures you never lose any speech, regardless of when the call ends.
ACS Call Recording Overview
Azure Speech to Text - REST API
Azure Speech-to-Text Capabilities

If this answer was helpful and pointed you in the right direction, please consider clicking "Accept Answer" and kindly Upvote — it may benefit other community members reading this thread. If you have any further questions, feel free to let us know.
Bhargavi Naragani 6,050 Reputation points Microsoft External Staff Moderator

2025-05-14T01:15:00.2133333+00:00

Hi Kino Matuguina,
Following up to see if the above answer was helpful. If this answers your query, do click Accept Answer and Yes, if you have any further query do let us know.

Share via

ACS Call Automation Misses Final Speech When Call Ends Immediately

0 additional answers

Your answer