TsgcAIVoiceTranslator translates speech while the user talks. It streams the microphone to OpenAI gpt-realtime-translate over a WebSocket, plays the translated speech as it arrives and reports live transcripts of what was said and of the translation.
A realtime speech to speech translator. Set the API key and the target language, call Start, speak. The model detects the source language, so there is nothing else to configure.
TsgcAIVoiceTranslator
Live speech to speech translation
Windows with microphone and speakers. Linux and macOS with your own audio.
Editions. TsgcAIVoiceTranslator ships in sgcAI and in sgcWebSockets Enterprise, the same editions as the realtime voice agent TsgcAIVoiceAgent. sgcAI is self-contained: it includes the sgcWebSockets Core runtime the component is built on. Available for Delphi 7 to Delphi 13 and C++ Builder.
Set the API key and the target language, assign the transcript events, call Start.
uses
sgcWebSocket_Types, sgcAI_VoiceTranslator;
procedure TForm1.StartTranslator;
begin
oTranslator := TsgcAIVoiceTranslator.Create(Self);
oTranslator.ApiKey := 'sk-...';
oTranslator.Language := 'es'; // speak any language, hear Spanish
oTranslator.NotifyEvents := neAsynchronous;
oTranslator.OnInputTranscript := DoInputTranscript;
oTranslator.OnOutputTranscript := DoOutputTranscript;
oTranslator.OnError := DoError;
oTranslator.Start; // microphone in, translated speech out
end;
procedure TForm1.DoInputTranscript(Sender: TObject; const aDelta: string);
begin
memoSource.Text := memoSource.Text + aDelta; // what was said
end;
procedure TForm1.DoOutputTranscript(Sender: TObject; const aDelta: string);
begin
memoTarget.Text := memoTarget.Text + aDelta; // the translation
end;
procedure TForm1.DoError(Sender: TObject; const aError: string);
begin
ShowMessage(aError);
end;
procedure TForm1.SwitchLanguage;
begin
oTranslator.Language := 'fr'; // applied to the running session
end;// includes: sgcWebSocket_Types.hpp, sgcAI_VoiceTranslator.hpp
void TForm1::StartVoiceTranslator()
{
oTranslator = new TsgcAIVoiceTranslator(this);
oTranslator->ApiKey = "sk-...";
oTranslator->Language = "es";
oTranslator->NotifyEvents = neAsynchronous;
oTranslator->OnInputTranscript = DoInputTranscript;
oTranslator->OnOutputTranscript = DoOutputTranscript;
oTranslator->OnError = DoError;
oTranslator->Start();
}
void __fastcall TForm1::DoInputTranscript(TObject *Sender, const String aDelta)
{
memoSource->Text = memoSource->Text + aDelta;
}
void __fastcall TForm1::DoOutputTranscript(TObject *Sender, const String aDelta)
{
memoTarget->Text = memoTarget->Text + aDelta;
}
void __fastcall TForm1::DoError(TObject *Sender, const String aError)
{
ShowMessage(aError);
}Turn the devices off to translate audio from a file, a phone line or a network stream. Push PCM16 at 24 kHz mono with SendAudio and take the translated speech from OnOutputAudio.
procedure TForm1.StartCustomAudio(const aPCM16: TBytes);
begin
oTranslator.AudioCapture := False; // no microphone
oTranslator.AudioPlayback := False; // no speakers
oTranslator.OnOutputAudio := DoOutputAudio;
oTranslator.Start;
oTranslator.SendAudio(aPCM16); // PCM16, 24 kHz, mono
end;
procedure TForm1.DoOutputAudio(Sender: TObject; const aPCM16: TBytes);
begin
// translated speech, PCM16 24 kHz mono, in order
end;The members you reach for most often.
Start connects, waits for the session, sends the target language and opens the devices. Stop flushes the pending audio and closes the session. Active does the same from the Object Inspector.
The target language code, en by default. Assign a new value while the session runs and the change is sent to the model at once, no reconnect needed.
ApiKey, Model (gpt-realtime-translate) and BaseURL. SafetyIdentifier sends a hashed user id in the OpenAI-Safety-Identifier header.
AudioCapture and AudioPlayback use the microphone and the speakers. Set them False and feed SendAudio with PCM16 at 24 kHz mono instead.
OnInputTranscript streams the text of what was said, OnOutputTranscript the text of the translation, both as deltas.
OnSessionCreated, OnOutputAudio with each translated audio chunk, OnSessionClosed and OnError. NotifyEvents selects asynchronous events for GUI applications or no synchronization for services.
Platforms. The microphone and speaker path uses the Windows audio API. On Linux and macOS set AudioCapture and AudioPlayback to False and handle the audio yourself with SendAudio and OnOutputAudio. With the devices left on, Start reports through OnError that no audio device is available on that platform.
Authoritative sources for the service this component uses.
| Online HelpFull API reference and usage guide for the sgcAI components. | Open | |
| OpenAI TranslatorRecord, translate and speak one segment at a time with TsgcAIOpenAITranslator. | Open | |
| All sgcAI ComponentsBrowse the full feature matrix of the sgcAI components. | Open | |
| Download Free TrialIncludes the Voice Translator demo. Bring your own OpenAI key. | Open | |
| PricingSingle, Team and Site licenses, full source code. | Open |