Method run

src/framework/modules/hift_vocoder.cpp:710–737

    }

    HiftVocoderOutput run(
        const std::vector<float> & mel,
        int64_t frames,
        uint64_t seed,
        uint64_t prior_noise_values,
        const std::vector<float> * source_random_values) {
        std::lock_guard<std::mutex> lock(mutex_);
        const auto f0 = predict_f0_locked(mel, frames);
        const int64_t scale = upsample_scale(weights_->config);
        const auto f0_upsampled = nearest_upsample_f0(f0, frames, scale);
        const auto source = make_sine_source(*weights_, f0_upsampled, seed, prior_noise_values, source_random_values);
        std::vector<float> window;
        int64_t stft_frames = 0;
        const auto source_stft = source_stft_bct(*weights_, source, stft_frames, window);
        if (backend_runner_ != nullptr && !backend_runner_->supports(*weights_, frames, stft_frames)) {
            backend_runner_.reset();
        }
        if (backend_runner_ == nullptr) {
            backend_runner_ = std::make_unique<BackendRunner>(*weights_, frames, stft_frames);
        }
        const auto post = backend_runner_->run(mel, frames, source_stft, stft_frames);
        HiftVocoderOutput out;
        out.waveform = waveform_from_post(*weights_, post, stft_frames, window);
        out.batch = 1;
        out.samples = static_cast<int64_t>(out.waveform.size());
        out.sample_rate = weights_->config.sampling_rate;
        return out;
    }

    void release_runtime_cache() {
        std::lock_guard<std::mutex> lock(mutex_);