Real-time SDK — Go

Stream audio to MyVocal Speech to Text from Go 1.22+ with package sttrealtime.

go get github.com/MyVocal-AI/myvocal-stt-realtime-sdks/go@v1.0.0

Go 1.22 or later. Dependency: github.com/gorilla/websocket. Served by the Go module proxy from the go/v1.0.0 tag of MyVocal-AI/myvocal-stt-realtime-sdks.

Stream a file

import sttrealtime "github.com/MyVocal-AI/myvocal-stt-realtime-sdks/go"

client, err := sttrealtime.NewClient(sttrealtime.Config{AccessKey: os.Getenv("MYVOCAL_ACCESS_KEY")})
if err != nil { log.Fatal(err) }
defer client.Close()

ctx := context.Background()
session, err := client.Connect(ctx, sttrealtime.ConnectOptions{
    Request: &sttrealtime.SessionRequest{Title: "meeting", LanguageHint: "en",
        Options: &sttrealtime.SessionOptions{InputEncoding: "pcm_s16le_16000", SegmentCommitMode: "vad"}},
    Handlers: map[string]func(sttrealtime.Event){
        sttrealtime.EventTranscriptFinal: func(e sttrealtime.Event) { fmt.Println("final:", e.Text()) },
    },
})
if err != nil { log.Fatal(err) }
defer session.Close()                                    // local resources only

if err := session.SendFile(ctx, "speech.wav", true); err != nil { log.Fatal(err) } // paced
result, err := session.Finish(ctx)
if err != nil { log.Fatal(err) }
fmt.Println(result.Status, result.TranscriptionID)

Every call takes a context.Context; cancel it to stop waiting. Optional booleans and numbers use sttrealtime.Bool, Int and Float, so false and 0 are sent as given and nil is omitted.

Byte chunks and events

session, err := client.Connect(ctx, sttrealtime.ConnectOptions{Request: req, EventChannel: true,
    OnCallbackError: func(err error, e sttrealtime.Event) { log.Println("handler failed on", e.Type) }})
go func() {
    for e := range session.Events() { /* typed: e.Type, e.Text(), e.Usage(), e.Notice() ... */ }
}()
for chunk := range captured {                            // any size; not paced again
    if err := session.SendAudio(ctx, chunk); err != nil { break }
}
result, err := session.Finish(ctx)

Commit, pause and resume

err = session.Commit(ctx)                        // line boundary; same epoch
err = session.Pause(ctx, sttrealtime.ViaWS)      // or ViaREST
err = session.Resume(ctx, sttrealtime.ViaREST)   // new epoch

After a disconnect

result, err := session.Finish(ctx)
var unconfirmed *sttrealtime.FinishUnconfirmedError
switch {
case err == nil && result.ViaREST:
    log.Println("closed out:", result.Status, result.CloseoutReason)
case errors.As(err, &unconfirmed):
    view, gerr := client.GetSession(ctx, unconfirmed.SessionID)   // later
    if gerr == nil && !view.IsTerminal() { view, gerr = client.FinishSession(ctx, unconfirmed.SessionID, nil) }
}

Config

Field	Default	Meaning
BaseURL	https://api.myvocal.ai	http/https; the stream uses ws/wss
Timeouts	see behaviour guide	zero fields use defaults
FrameMs, MaxBufferedMs	100, 30000	frame size and the bound of unacknowledged audio
PingInterval	20 s	client WebSocket ping
EventQueueSize	10000	bound of events waiting for handlers
HTTPClient, Proxy (NoProxy, ProxyURL), TLSConfig	environment proxy, system roots	a borrowed client is never closed
DisableCloseoutOnDisconnect	false	turn off the automatic REST close-out after an unexpected disconnect (Finish still closes out)

Errors

Use errors.Is with ErrInputInvalid, ErrAuthentication, ErrInsufficientCharacters, ErrNotFound, ErrIdempotencyConflict, and errors.As with *APIError (Code, ErrorCode, Retryable, RequestID), *AudioFormatError, *SessionStateError, *ConnectionLostError, *BackpressureTimeoutError, *FinishUnconfirmedError (SessionID) and *TransportError.