voskspeechtotextdevice.cpp

/*
  SPDX-FileCopyrightText: 2023-2025 Laurent Montel <montel@kde.org>
 
  SPDX-License-Identifier: GPL-2.0-or-later
  based on VoiceAssistant plugin code
*/
 
#include "voskspeechtotextdevice.h"
#include "libvoskspeechtotext_debug.h"
#if HAVE_VOSK_API_SUPPORT
#include "vosk_api.h"
#endif
#include <QJsonDocument>
 
VoskSpeechToTextDevice::VoskSpeechToTextDevice(QObject *parent)
    : QIODevice{parent}
{
    if (!open(QIODevice::ReadWrite)) {
        qCWarning(LIBVOSKSPEECHTOTEXT_LOG) << "Impossible to open VoskSpeechToTextDevice";
#if HAVE_VOSK_API_SUPPORT
        vosk_set_log_level(-1);
#endif
    }
}
 
VoskSpeechToTextDevice::~VoskSpeechToTextDevice()
{
#if HAVE_VOSK_API_SUPPORT
    vosk_recognizer_free(mRecognizer);
    vosk_model_free(mModel);
#endif
}
 
bool VoskSpeechToTextDevice::available() const
{
#if HAVE_VOSK_API_SUPPORT
    return true;
#else
    return false;
#endif
}
 
bool VoskSpeechToTextDevice::isAsking() const
{
    return mIsAsking;
}
 
void VoskSpeechToTextDevice::setAsking(bool asking)
{
    if (mIsAsking != asking) {
        mIsAsking = asking;
        Q_EMIT askingChanged();
    }
}
 
bool VoskSpeechToTextDevice::initialize(VoskSpeechToTextDeviceInfo &&info)
{
#if HAVE_VOSK_API_SUPPORT
    mModel = vosk_model_new(QString(info.modelDir + info.formattedLang).toUtf8().constData());
    if (mModel) {
        mRecognizer = vosk_recognizer_new(mModel, info.sampleRate);
    }
 
    if (!mModel || !mRecognizer) {
        return false;
    }
#endif
    return true;
}
 
void VoskSpeechToTextDevice::clear()
{
#if HAVE_VOSK_API_SUPPORT
    if (mRecognizer) {
        vosk_recognizer_reset(mRecognizer);
    }
#endif
}
 
qint64 VoskSpeechToTextDevice::readData(char *data, qint64 maxlen)
{
    Q_UNUSED(data);
    return maxlen;
}
 
qint64 VoskSpeechToTextDevice::writeData(const char *data, qint64 len)
{
#if HAVE_VOSK_API_SUPPORT
    if (vosk_recognizer_accept_waveform(mRecognizer, data, (int)len)) {
        parseText(vosk_recognizer_result(mRecognizer));
    } else {
        parsePartial(vosk_recognizer_partial_result(mRecognizer));
    }
#else
    Q_UNUSED(data);
#endif
    return len;
}
 
void VoskSpeechToTextDevice::parseText(const char *json)
{
    const QJsonDocument obj = QJsonDocument::fromJson(json);
    QString text = obj[QStringLiteral("text")].toString();
 
    if (text.isEmpty())
        return;
    else if (mIsAsking) {
        Q_EMIT result(text);
        return;
    }
 
    text.append(u' ');
 
    if (!text.contains(mWakeWord)) {
        if (!mIsListiningBecauseOfWakeWord)
            return;
 
        Q_EMIT falsePositiveWakeWord();
        mIsListiningBecauseOfWakeWord = false;
        return;
    }
 
    text = text.mid(text.indexOf(mWakeWord) + mWakeWord.size());
    text = text.trimmed();
 
    Q_EMIT result(text);
    qDebug() << "[debug] Text:" << text;
    Q_EMIT doneListening();
}
 
void VoskSpeechToTextDevice::parsePartial(const char *json)
{
    const QJsonDocument obj = QJsonDocument::fromJson(json);
    QString text = obj[QStringLiteral("partial")].toString();
    if (text.isEmpty())
        return;
    text.append(u' ');
 
    if (text.contains(mWakeWord)) {
        Q_EMIT wakeWordDetected();
        text = text.mid(text.indexOf(mWakeWord) + mWakeWord.size());
        mIsListiningBecauseOfWakeWord = true;
    } else if (mIsListiningBecauseOfWakeWord) {
        Q_EMIT falsePositiveWakeWord();
        mIsListiningBecauseOfWakeWord = false;
        return;
    } else if (!mIsAsking)
        return;
 
    Q_EMIT result(text);
}
 
QDebug operator<<(QDebug d, const VoskSpeechToTextDevice::VoskSpeechToTextDeviceInfo &t)
{
    d.space() << "sampleRate" << t.sampleRate;
    d.space() << "modelDir" << t.modelDir;
    d.space() << "formattedLang" << t.formattedLang;
    return d;
}
 
#include "moc_voskspeechtotextdevice.cpp"