-- SPDX-FileCopyrightText: © 2026 Vladimir Zorin <vladimir@deviant.guru>
-- SPDX-License-Identifier: LicenseRef-OWL-1.0-or-later
-- Licensed under OWL v1.0+. See LICENSE.

-- Manual audition harness for sound.speech — the tool for tuning the phoneme
-- tables, prosody and voice presets by ear. NOT part of run_tests.bash.
--
--   lilush tests/sound/manual/speak.lua                     play all presets
--   lilush tests/sound/manual/speak.lua computer            one preset
--   lilush tests/sound/manual/speak.lua computer "text..."  custom line
--   lilush tests/sound/manual/speak.lua --wav [dir]         dump WAVs instead
--                                                           (default ./speech_demo)
--
-- Playback needs a real PCM device (see smoke.lua: card/device via
-- LILUSH_SND_CARD / LILUSH_SND_DEVICE). The WAV path needs no hardware —
-- listen off-box, or feed a spectrogram tool when tuning formants.

local sound = require("sound")
-- fall back to a direct require when running against a binary built before
-- the speech module existed (dev tree via ./?.lua)
local speech = sound.speech or require("sound.speech")

local LINES = {
	"Docking request granted. Proceed to pad 3.",
	"Warning. Hull temperature critical.",
	"You have arrived. Welcome to Haven station.",
}

-- ── tiny WAV writer (PCM s16, mono) ──────────────────────────────────

local u16 = function(v)
	return string.char(v % 256, math.floor(v / 256) % 256)
end

local u32 = function(v)
	local b1 = v % 256
	local b2 = math.floor(v / 256) % 256
	local b3 = math.floor(v / 65536) % 256
	local b4 = math.floor(v / 16777216) % 256
	return string.char(b1, b2, b3, b4)
end

local write_wav = function(path, buf, rate)
	local n = buf:frames()
	local parts = {}
	for i = 1, n do
		local s = buf:get(i, 1)
		if s > 1 then
			s = 1
		elseif s < -1 then
			s = -1
		end
		local v = math.floor(s * 32767 + 0.5)
		if v < 0 then
			v = v + 65536
		end
		parts[i] = string.char(v % 256, math.floor(v / 256) % 256)
	end
	local data = table.concat(parts)
	local f = assert(io.open(path, "wb"))
	f:write("RIFF", u32(36 + #data), "WAVE")
	f:write("fmt ", u32(16), u16(1), u16(1), u32(rate), u32(rate * 2), u16(2), u16(16))
	f:write("data", u32(#data), data)
	f:close()
end

-- ── argument parsing ─────────────────────────────────────────────────

local wav_dir
local preset_arg, text_arg
local i = 1
while arg[i] do
	if arg[i] == "--wav" then
		wav_dir = arg[i + 1] and not arg[i + 1]:match("^%-") and arg[i + 1] or "./speech_demo"
		if wav_dir == arg[i + 1] then
			i = i + 1
		end
	elseif not preset_arg then
		preset_arg = arg[i]
	else
		text_arg = (text_arg and text_arg .. " " or "") .. arg[i]
	end
	i = i + 1
end

local presets = {}
if preset_arg then
	assert(speech.voices[preset_arg], "unknown preset: " .. tostring(preset_arg))
	presets = { preset_arg }
else
	for name in pairs(speech.voices) do
		presets[#presets + 1] = name
	end
	table.sort(presets)
end
local lines = text_arg and { text_arg } or LINES

-- ── WAV mode (no hardware needed) ────────────────────────────────────

if wav_dir then
	os.execute("mkdir -p '" .. wav_dir .. "'")
	for _, name in ipairs(presets) do
		for li, line in ipairs(lines) do
			local buf, info = speech.say(line, { voice = name })
			assert(buf, info)
			local path = ("%s/%s_%d.wav"):format(wav_dir, name, li)
			write_wav(path, buf, info.rate)
			print(("%-32s %.2fs  %q"):format(path, info.dur, line))
		end
	end
	return
end

-- ── playback mode ────────────────────────────────────────────────────

local lev = require("lev")

local card = tonumber(os.getenv("LILUSH_SND_CARD")) or 0
local devno = tonumber(os.getenv("LILUSH_SND_DEVICE")) or 0
local dev, err = sound.open({ card = card, device = devno, rate = 48000, channels = 2 })
if not dev then
	io.stderr:write(("could not open audio device hw:%d,%d: %s\n"):format(card, devno, tostring(err)))
	io.stderr:write("(no device? use --wav to dump files instead)\n")
	os.exit(1)
end
local mx = sound.mixer({ channels = dev:channels() })

lev.run(function()
	local token = lev.cancel_token()
	sound.run(dev, mx, { cancel = token })
	for _, name in ipairs(presets) do
		for _, line in ipairs(lines) do
			local buf, info = speech.say(line, { voice = name, rate = dev:rate() })
			assert(buf, info)
			print(("%-10s %q"):format(name, line))
			io.stdout:flush()
			mx:play(buf, { gain = 0.9 })
			lev.sleep(info.dur + 0.4)
		end
	end
	token:cancel()
	lev.sleep(0.2)
	print("done.")
end)