talk/internal/provider/osbuiltin.go

package provider

import (
	"bytes"
	"context"
	"encoding/base64"
	"encoding/binary"
	"errors"
	"fmt"
	"os"
	"os/exec"
	"runtime"
	"unicode/utf16"
)

// OSBuiltin speaks through the TTS runtime built into the operating system.
// It is the chain's provider of last resort: present everywhere, without a
// key and without network, but with the plainest voice.
//
// Platform mechanisms (selected at runtime via runtime.GOOS so that one
// binary compiles for all targets; every route is pure exec without a
// shell):
//
//	Linux:   Synthesize via espeak/espeak-ng (--stdout → WAV bytes).
//	         If both are missing, Synthesize returns ErrNoByteOutput and
//	         SpeakDirect plays via spd-say (speech-dispatcher); according
//	         to its own --help output spd-say can only emit audio itself
//	         through the daemon, not write it to a file or a process
//	         (verified on the dev host 2026-10-18).
//	macOS:   Synthesize via say (-o temporary WAV file,
//	         --data-format=LEI16@22050); voices de→Anna, en→Samantha
//	         (preinstalled).
//	Windows: Synthesize via PowerShell System.Speech into a temporary WAV
//	         file (SetOutputToWaveFile); language via SelectVoiceByHints
//	         with fallback to the system voice.
//
// Missing binaries/runtimes lead to Available()==false or clear errors,
// never to panics.
type OSBuiltin struct{}

// Name implements Provider.
func (OSBuiltin) Name() string { return "osbuiltin" }

// Available implements Provider: true if at least one platform mechanism is
// present.
func (OSBuiltin) Available(ctx context.Context) bool {
	switch runtime.GOOS {
	case "darwin":
		_, err := exec.LookPath("say")
		return err == nil
	case "windows":
		_, err := exec.LookPath("powershell")
		return err == nil
	default: // linux and other unix-like systems
		for _, bin := range []string{"espeak", "espeak-ng", "spd-say"} {
			if _, err := exec.LookPath(bin); err == nil {
				return true
			}
		}
		return false
	}
}

// Describe names the platform mechanism found, for diagnosis
// (`talk doctor`): e.g. "espeak-ng (audio bytes)", "spd-say (direct speech
// only; install espeak-ng for audio bytes)", "say", "powershell (SAPI)",
// or what is missing.
func (OSBuiltin) Describe() string {
	switch runtime.GOOS {
	case "darwin":
		if _, err := exec.LookPath("say"); err == nil {
			return "say (macOS)"
		}
		return "say not found"
	case "windows":
		if _, err := exec.LookPath("powershell"); err == nil {
			return "powershell (SAPI System.Speech)"
		}
		return "powershell not found"
	default:
		for _, bin := range []string{"espeak-ng", "espeak"} {
			if _, err := exec.LookPath(bin); err == nil {
				return bin + " (audio bytes)"
			}
		}
		if _, err := exec.LookPath("spd-say"); err == nil {
			return "spd-say (direct speech only; install espeak-ng for audio bytes)"
		}
		return "neither espeak-ng/espeak nor spd-say found (install espeak-ng)"
	}
}

// Synthesize implements Provider. Return format: RIFF/WAVE (PCM), playable
// through every sink of the audio package.
func (p OSBuiltin) Synthesize(ctx context.Context, text, lang string) ([]byte, error) {
	if text == "" {
		return nil, fmt.Errorf("osbuiltin: empty text")
	}
	switch runtime.GOOS {
	case "darwin":
		return p.synthSay(ctx, text, lang)
	case "windows":
		return p.synthSAPI(ctx, text, lang)
	default:
		return p.synthEspeak(ctx, text, lang)
	}
}

// SpeakDirect implements DirectSpeaker: direct, blocking output through the
// OS TTS with its own playback. Required on Linux without espeak; not
// needed on macOS/Windows (Synthesize returns WAV) and a clear error there.
func (p OSBuiltin) SpeakDirect(ctx context.Context, text, lang string) error {
	if text == "" {
		return fmt.Errorf("osbuiltin: empty text")
	}
	switch runtime.GOOS {
	case "darwin", "windows":
		return fmt.Errorf("osbuiltin: SpeakDirect not needed on %s; Synthesize returns WAV, play it through a sink", runtime.GOOS)
	default:
		if _, err := exec.LookPath("spd-say"); err != nil {
			return fmt.Errorf("osbuiltin: spd-say not found (install speech-dispatcher, or espeak for audio bytes): %w", err)
		}
		cmd := exec.CommandContext(ctx, "spd-say", "-w", "-l", lang, text)
		if out, err := cmd.CombinedOutput(); err != nil {
			return fmt.Errorf("osbuiltin: spd-say failed: %w: %s", err, bytes.TrimSpace(out))
		}
		return nil
	}
}

// synthEspeak tries espeak, then espeak-ng. Both write WAV (RIFF/PCM) to
// stdout.
func (OSBuiltin) synthEspeak(ctx context.Context, text, lang string) ([]byte, error) {
	for _, bin := range []string{"espeak", "espeak-ng"} {
		if _, err := exec.LookPath(bin); err != nil {
			continue
		}
		cmd := exec.CommandContext(ctx, bin, "-v", lang, "--stdout", text)
		out, err := cmd.Output()
		if err != nil {
			return nil, fmt.Errorf("osbuiltin: %s failed: %w", bin, withStderr(err))
		}
		if len(out) == 0 {
			return nil, fmt.Errorf("osbuiltin: %s returned no audio data", bin)
		}
		return out, nil
	}
	return nil, fmt.Errorf("osbuiltin: no espeak/espeak-ng for audio bytes found on %s: %w", runtime.GOOS, ErrNoByteOutput)
}

// synthSay produces WAV via the macOS say command into a temporary file.
// --data-format=LEI16@22050 forces 16-bit PCM; the .wav extension forces
// the WAVE container.
func (OSBuiltin) synthSay(ctx context.Context, text, lang string) ([]byte, error) {
	if _, err := exec.LookPath("say"); err != nil {
		return nil, fmt.Errorf("osbuiltin: say not found: %w", err)
	}
	tmp, err := os.CreateTemp("", "talk-say-*.wav")
	if err != nil {
		return nil, fmt.Errorf("osbuiltin: temporary file: %w", err)
	}
	name := tmp.Name()
	tmp.Close()
	defer os.Remove(name)

	cmd := exec.CommandContext(ctx, "say", "-v", sayVoice(lang),
		"--data-format=LEI16@22050", "-o", name, text)
	if out, err := cmd.CombinedOutput(); err != nil {
		return nil, fmt.Errorf("osbuiltin: say failed: %w: %s", err, bytes.TrimSpace(out))
	}
	data, err := os.ReadFile(name)
	if err != nil {
		return nil, fmt.Errorf("osbuiltin: reading say output: %w", err)
	}
	return data, nil
}

// sayVoice maps languages to preinstalled macOS voices.
func sayVoice(lang string) string {
	if lang == "en" {
		return "Samantha" // en-US
	}
	return "Anna" // de-DE
}

// synthSAPI produces WAV via the Windows SAPI (System.Speech) through
// PowerShell. The script is passed as -EncodedCommand (Base64/UTF-16LE) and
// the text inside it is passed base64-decoded, so there is no quoting
// problem with special characters. If the requested language is not
// installed, SAPI falls back to the system voice (try/catch).
func (OSBuiltin) synthSAPI(ctx context.Context, text, lang string) ([]byte, error) {
	if _, err := exec.LookPath("powershell"); err != nil {
		return nil, fmt.Errorf("osbuiltin: powershell not found: %w", err)
	}
	tmp, err := os.CreateTemp("", "talk-sapi-*.wav")
	if err != nil {
		return nil, fmt.Errorf("osbuiltin: temporary file: %w", err)
	}
	name := tmp.Name()
	tmp.Close()
	defer os.Remove(name)

	script := fmt.Sprintf(`$ErrorActionPreference = 'Stop'
Add-Type -AssemblyName System.Speech
$s = New-Object System.Speech.Synthesis.SpeechSynthesizer
try {
  $culture = New-Object System.Globalization.CultureInfo('%s')
  $s.SelectVoiceByHints([System.Speech.Synthesis.VoiceGender]::Female, [System.Speech.Synthesis.VoiceAge]::Adult, 0, $culture)
} catch { }
$s.SetOutputToWaveFile('%s')
$s.Speak([System.Text.Encoding]::UTF8.GetString([Convert]::FromBase64String('%s')))
$s.Dispose()
`, sapiCulture(lang), name, base64.StdEncoding.EncodeToString([]byte(text)))

	cmd := exec.CommandContext(ctx, "powershell", "-NoProfile", "-STA", "-EncodedCommand", psEncode(script))
	if out, err := cmd.CombinedOutput(); err != nil {
		return nil, fmt.Errorf("osbuiltin: SAPI failed: %w: %s", err, bytes.TrimSpace(out))
	}
	data, err := os.ReadFile(name)
	if err != nil {
		return nil, fmt.Errorf("osbuiltin: reading SAPI output: %w", err)
	}
	return data, nil
}

// sapiCulture maps languages to SAPI culture identifiers.
func sapiCulture(lang string) string {
	if lang == "en" {
		return "en-US"
	}
	return "de-DE"
}

// psEncode encodes a PowerShell script for -EncodedCommand
// (UTF-16LE + Base64).
func psEncode(script string) string {
	runes := utf16.Encode([]rune(script))
	buf := make([]byte, len(runes)*2)
	for i, r := range runes {
		binary.LittleEndian.PutUint16(buf[i*2:], r)
	}
	return base64.StdEncoding.EncodeToString(buf)
}

// withStderr enriches an exec error with the process's collected stderr
// output, if any.
func withStderr(err error) error {
	var exitErr *exec.ExitError
	if errors.As(err, &exitErr) && len(exitErr.Stderr) > 0 {
		return fmt.Errorf("%w: %s", err, bytes.TrimSpace(exitErr.Stderr))
	}
	return err
}