stt

func AsTranscriber ¶ added in v0.61.28

func AsTranscriber(p STTProvider, opts ...TranscriberOption) speechkit.Transcriber

AsTranscriber adapts an STTProvider to the kernel's speechkit.Transcriber interface, so hosts can hand any provider straight to dictation.NewRuntime or a TranscriptionWorker without hand-writing the mechanical Result-to-Transcript adapter. The per-call language overrides the base option language when non-empty; when both are empty the request carries no language at all — the adapter never invents a default, because language selection (including multilanguage) is an explicit caller decision.

The device app keeps its own richer internal adapter (vocabulary bias, customization, routing); this bridge covers the plain provider-to-kernel path for library consumers.

package main

import (
	"context"
	"fmt"
	"log"

	"github.com/kombifyio/SpeechKit/pkg/speechkit"
	"github.com/kombifyio/SpeechKit/pkg/speechkit/stt"
)

// echoProvider is a minimal stt.STTProvider. Real hosts use the providers in
// the stt subpackages (local, deepgram, openai, ...) or their own backend.
type echoProvider struct{ name string }

func (p echoProvider) Name() string                 { return p.name }
func (p echoProvider) Health(context.Context) error { return nil }
func (p echoProvider) Transcribe(_ context.Context, _ []byte, opts stt.TranscribeOpts) (*stt.Result, error) {
	return &stt.Result{Text: "hello from " + p.name, Language: opts.Language, Provider: p.name}, nil
}

func main() {
	var transcriber speechkit.Transcriber = stt.AsTranscriber(
		echoProvider{name: "echo"},
		stt.WithTranscribeOpts(stt.TranscribeOpts{Language: "de"}),
	)

	wav := speechkit.PCMToWAV(make([]byte, 16000*2))
	transcript, err := transcriber.Transcribe(context.Background(), wav, 1.0, "")
	if err != nil {
		log.Fatal(err)
	}
	fmt.Println(transcript.Text, transcript.Language, transcript.Provider)

	transcript, _ = transcriber.Transcribe(context.Background(), wav, 1.0, "en")
	fmt.Println(transcript.Language)
}

Output:
hello from echo de echo
en

func BaseCapabilities ¶ added in v0.64.3

func BaseCapabilities() []speechkit.Capability

BaseCapabilities are reported by every speech-to-text provider. A fresh slice is returned each call so callers may append without aliasing.

func EnsureTranscriptionWAV ¶ added in v0.64.3

func EnsureTranscriptionWAV(raw []byte) []byte

EnsureTranscriptionWAV wraps raw PCM in a WAV header when it does not already carry one, because every HTTP transcription endpoint expects a container rather than bare samples.

func EnvSecretResolver ¶ added in v0.67.14

func EnvSecretResolver(name string) string

EnvSecretResolver reads secrets from the process environment. It is the default every provider falls back to when no resolver is configured.

func FirstNonEmptyTrimmed ¶ added in v0.64.3

func FirstNonEmptyTrimmed(values ...string) string

FirstNonEmptyTrimmed returns the first value that is not blank after trimming, or "". Provider adapters use it to layer a request override over a provider default over a package default.

func IsMultilanguage ¶ added in v0.53.0

func IsMultilanguage(language string) bool

IsMultilanguage reports whether language asks for multilanguage rather than a specific locale. Empty, "auto" and "multi" all mean the same thing: do not pin. The settings UI offers "auto" and "multi" as separate labels, so both have to resolve here or the literal token reaches providers that reject it.

func IsWAV ¶ added in v0.64.3

func IsWAV(raw []byte) bool

IsWAV reports whether raw starts with a RIFF/WAVE header.

func IsWebSocketClose ¶ added in v0.64.3

func IsWebSocketClose(err error) bool

IsWebSocketClose reports whether err is a WebSocket close, as opposed to a transport failure. Streaming providers treat the two differently: a close is the end of a session, a transport failure is worth surfacing.

func MaxSpeakers ¶ added in v0.64.3

func MaxSpeakers(opts speaker.Options) int

MaxSpeakers is the upper-bound counterpart of MinSpeakers.

func MinSpeakers ¶ added in v0.64.3

func MinSpeakers(opts speaker.Options) int

MinSpeakers and MaxSpeakers collapse the three speaker-count knobs into the single lower and upper bound a provider request carries. An exact expected count wins over the range.

func PCM16FromWAV ¶ added in v0.64.3

func PCM16FromWAV(audio []byte) (pcm []byte, sampleRate, channels int, ok bool)

PCM16FromWAV extracts raw PCM16 sample data plus the sample rate and channel count from a LINEAR16 RIFF/WAVE byte slice. It returns ok=false when the input is not a 16-bit PCM WAV, in which case callers should treat the bytes as already-raw PCM and fall back to a default rate.

Providers that must declare the sample rate explicitly (e.g. Google STT v1 speech:recognize) rely on this to avoid sending a wrong hard-coded rate, which the provider rejects with HTTP 400 on a mismatch.

func ProviderNameFromProfileID ¶ added in v0.64.3

func ProviderNameFromProfileID(profileID string) string

ProviderNameFromProfileID extracts the bare provider name from a model_selection profile id, or "" when the id names a specific model.

func ToTranscript ¶ added in v0.61.28

func ToTranscript(r *Result, durationSecs float64) speechkit.Transcript

ToTranscript maps a provider Result to a speechkit.Transcript. The mapping is mechanical: text, language, provider, model, confidence, per-word confidences, and speaker diarization pass through 1:1. Duration comes from the Result when the provider reported one; otherwise durationSecs (the caller-measured audio length) fills in. A nil Result yields a zero Transcript.