Hello.
I am using channel.play to play a prerecorded audio using OpenAI’s function calling:
const introPlayback = await ariClient.channels.play({
*channelId,*
*media: 'sound:/var/lib/asterisk/sounds/intro_male_fast'*
*});*
*activePlaybacks.set(channelId, introPlayback);
Is there a method through which I can actively track the playback so that I can interrupt it during playback?
It returns an active playback that you can then control[1]. How you store the playback instance information in your ARI application is up to you. Right now it would seem as though you are storing it associated with the channel id.
If this doesn’t cover what you are asking, you’re going to need to be more explicit.
Thank you for your reply.
I managed to retrieve a unique PlaybackID which I’m getting from ARI which I am storing locally. I can actively track the Playback using that.
If I manage to kill the playback using the PlaybackID, I think I can interrupt it.
Is there any method you can suggest for deleting it or stopping it?
You stop it by calling DELETE on the playback. How you do that in your language is on you.
I have been trying that approach. And I found the following:
Each Asterisk channel has a read and write direction for media?
When we use ariClient.channels.play({ channelId, media: ‘sound:intro’ }) to play the audio, the channel is occupied for playing the audio recording.
Which means that it is not listening to the user audio during playback as Asterisk temporarily suspends frame forwarding to ARI, until playback has ended.
You can confirm these findings i.e. ARI’s nature during channels.play
And if this is the case, then I think using channels.play for playing an audio recording, I would not be able to produce an interruption in the audio.
Is there any method I can try besides channels.play for playing a recording for my use case i.e., allowing real-time barge-in or interruption by the caller?
Because it seems that while channels.play() is running, the channel stops forwarding inbound RTP frames (caller audio) to ARI.
Correct, this is how playback works.
If you want to have the channel continue to do other things while playing audio to it, you would use a snoop channel[1] to whisper into it and play back a recording on the snoop channel. The original channel would then be free to do as it wishes. If this doesn’t do what you want, then you’ll need to be much more specific and detailed about what you are attempting to achieve instead of bits and pieces.
Does the snoop channel have an audio playback function I can use like channels.play? To play sounds from the sounds directory in Asterisk?
Snoop channels are a channel. You use them as a channel. So you would call the same channels.play
Okay, thank you for the help.
I will try it out and update here.
So I tried using the snoop channel, but the problem is that the audio being played as a recording is interpreted by OpenAI as user input, so it gets interrupted immediately and never finishes playing.
And I apologize about this, I did not read your entire message before. Basically what I’m trying to develop is using OpenAI’s Realtime to play Prerecorded audios using ARI through OpenAI’s function calling. I’ve implemented real-time speech detection (via WebSocket) to detect when the caller starts speaking so I can interrupt ongoing audio playback (similar to “barge-in”).
A Snoop channel whispers audio in the direction you specify when you create it. If the channel you are snooping on is not OpenAI, then you’ve specified the direction incorrectly when creating the snoop channel. Both=Whispered into audio going to the channel and audio coming from the channel. To=Whispered into audio going to the channel. From=Whispered into audio coming from the channel.
Are you using AI or an LLM to write the code to implement this?
Thank you for specifying the directions for the Snoop channel. I will look into this and update here.
Yes, for Asterisk, I am using AI as this is uncharted territory for me.
We find that most people who try to use AI to code for Asterisk end up with incorrect solutions.
Indeed, it was the reason I asked - because that Snoop direction thing has come up 3 or 4 times, from people using AI to implement.
That makes sense. I have been primarily using AI as a learning tool for this rather than a coding authority because I’m still getting to know Asterisk’s architecture.
This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.