The real-time transcriber should have an onclose callback so users can know why the connection was closed.

The wordSearch method has an overload without any words which doesn't make sense:

public WordSearchResponse wordSearch(String transcriptId) {

This overload should be removed, not generated.

Transcripts with longer token size time out with LeMUR functions.

Requests with longer files result in this error: Caused by: java.net.SocketTimeoutException: timeout

For context, here is the code used:

import com.assemblyai.api.AssemblyAI;
import com.assemblyai.api.resources.lemur.requests.*;
import com.assemblyai.api.resources.transcripts.types.*;
import java.util.List;

public final class App {
    public static void main(String[] args) {

        AssemblyAI client = AssemblyAI.builder()
                .apiKey("api-key")
                .build();

        String url = "https://storage.googleapis.com/aai-web-samples/meeting.mp3";

        Transcript transcript = client.transcripts().transcribe(url);

        String prompt = "Provide a brief summary of the transcript.";

        var params = LemurTaskParams.builder()
                .prompt(prompt)
                .transcriptIds(List.of(transcript.getId()))
                .build();

        var result = client.lemur().task(params);

        System.out.println(result.getResponse());

Although it is not stated in the repository, i am working under the assumption that the library is supposed to work on the widespead java 1.8 version.
Unfortunately it seems that the AssemblyAI.build() method throws a NoSuchMethodException when compiled and run on java 1.8

The issue disappears when running on JDK 21.

Is the library supposed to work on java 1.8, and if not, what is the minimum version required?

When providing the method with a webm audio file, the assemblyAI platfom immediately shows that the transcription failed with an error, however the method does not throw an exception, and hangs indefinitely.

See spec: https://github.com/AssemblyAI/assemblyai-api-spec/blob/a28083d778c569a836eb65fcd045d330610ac5ec/openapi.yml#L883
https://github.com/AssemblyAI/assemblyai-api-spec/blob/main/openapi.yml#L1410

There's only one value for the enum, which is why Fern may have tripped up.

The other SDKs have README's that have more code samples.
This SDK could benefit from having the same samples.

Node: https://github.com/AssemblyAI/assemblyai-node-sdk
Ruby: https://github.com/assemblyai/assemblyai-ruby-sdk
Python: https://github.com/AssemblyAI/assemblyai-python-sdk

While in our spec the name of this error is Error, Java seems to use the Exception suffix for exceptions, we should stick to that.

A customer is using some of our sample code from our docs.

AssemblyAI client = AssemblyAI.builder()
        .apiKey("")
        .build();

String audioUrl = "https://storage.googleapis.com/aai-web-samples/5_common_sports_injuries.mp3";

var params = TranscriptOptionalParams.builder()
        .speakerLabels(true)
        .build();

try {
    Transcript transcript = client.transcripts().transcribe(audioUrl, params);

    System.out.println(transcript.getText());

    transcript.getUtterances().ifPresent(utterances ->
            utterances.forEach(utterance ->
                    System.out.println("Speaker " + utterance.getSpeaker() + ": " + utterance.getText())
            )
    );
} catch (Exception exception) {
    log.error(exception.getMessage() + " " + exception.getCause());
}

They're getting the following error:

Cannot invoke "com.assemblyai.api.core.RequestOptions.getTimeout()" because "requestOptions" is null null

When I take a peek at the code, every API method accepts request options, but they're never checked for null, so every API method is broken.

It would be helpful for devs to have a generated API reference to find Java SDK specific details.

answerFormat is needed to pass a query for lemur().actionItems. But LemurBaseParams doesn't have answerFormat as a parameter.

Currently:

var response = client.lemur().actionItems(LemurBaseParams
                .builder()
                .transcriptIds(List.of(transcript.getId()))
                .context(LemurBaseParamsContext.of("A GitLab meeting to discuss logistic"))
                .build());

Desired:

var response = client.lemur().actionItems(LemurBaseParams
                .builder()
                .transcriptIds(List.of(transcript.getId()))
                .context(LemurBaseParamsContext.of("A GitLab meeting to discuss logistic"))
                .answerFormat("Bullet points")
                .build());

Currently, the RealtimeMessage.visit method will throw when it receives an unknown message type.
In the future, our realtime service may send new message types which shouldn't introduce a breaking change.
Therefore, the implementation needs to change to not throw an error, instead maybe have an "unknown" overload too.

https://github.com/AssemblyAI/assemblyai-java-sdk/blob/main/src/main/java/com/assemblyai/api/resources/realtime/types/RealtimeMessage.java#L33-L46

I think the current visitor pattern makes sense in other scenarios, but in receiving realtime messages, it needs to be modified.

Title says it all. I want to be able to add params for different languages and also be able to turn automatic language detection on. Would it be possible to add this functionality to this SDK?

onPartialTranscript is not invoking where as all the events are firing in onSessionStart

RealtimeTranscriber.builder()
                .apiKey(ASSEMBLY_AI_API_KEY)
                .onPartialTranscript(this::displayPartialMessage)
                .onFinalTranscript(this::displayFinalMessage)
                .onError((err) -> {
                    displayErrorMessage(err.getMessage());
                })
                .onSessionStart(System.err::println)
                .build();

Please help.

context should be able to handle type String as a input instead of LemurBaseParamsContext.

Current functionality:

var response = client.lemur().summary(LemurSummaryParams.builder()
                .transcriptIds(List.of(transcript.getId()))
                .context(LemurBaseParamsContext.of("A GitLab meeting to discuss logistic"))
                .answerFormat("TLDR").build());

When you use the realtime transcriber with an empty API key, the error is not passed to onError.

I am getting the following output in the console.

{
  "message_type" : "SessionBegins",
  "error" : "Not authorized"
}

and

{
  "session_id" : "ab0ae487-ed4d-4d24-b6c3-65a701df20f6",
  "expires_at" : "2024-01-18T21:31:50.915638Z",
  "message_type" : "SessionBegins"
}
{
  "message_type" : "SessionBegins",
  "error" : "Received invalid request. Please check the documentation for the correct request schemas."
}

This is only in the Java SDK, so it's not how the service is sending the message to the client, but how the Java SDK is deserializing the message.

The SDK should support Java 8, but based on a recent customer interaction, the SDK is using some APIs that are not in Java 8.
Specifically the Map.of method.

Closing the real-time transcriber should send the terminate session message and wait for session termination.
Between the request for termination and confirmation for termination, partial and final transcripts will continue to come through.

The FilesClient and PollingTranscriptsClient should support streams so users can stream files from disk, network, etc.

Missing contentSafetyConfidence parameter in TranscriptOptionalParams class.

Desired usage example:

var params = TranscriptOptionalParams.builder()
             .contentSafety(true)
             .contentSafetyConfidence(60)
             .build();

Parameter: contentSafetyConfidence
Type: integer
Description: The confidence threshold for content moderation. Values must be between 25 and 100.

Docs Usage: https://www.assemblyai.com/docs/audio-intelligence/content-moderation#adjust-the-confidence-threshold

Word search only accepts a single parameter in addition to the ID, so it's a little ridiculous to use a builder for this.
It's fine to keep a builder for future expansion, but can we have an overload that takes a single string or a list of strings?

Currently:
client.transcripts().wordSearch(transcript.getId(), WordSearchParams.builder().words(word).build());

Desired:

client.transcripts().wordSearch(transcript.getId(), word);