Awtomatikong salin · Orihinal sa Ingles

Parakeet TDT — Android

Parakeet TDT v3 run sa Android, nakapaloob Linux, at Windows sa pamamagitan ONNX Runtime. Ang modelo ay nahahati sa mga talaksang grapiko at decoder-joint, kapuwa INT8 Inuri, na may TDT na masakim na pagsuporta 114 mga wika at isang 8192-Tome BPE bokabularyo.

Modelong mga Bida

Sawi ang paglikha ng talaksan Laki
parakeet-encoder-int8.onnx840 MB
parakeet-decoder-joint-int8.onnx51 MB
vocab.json<1 MB

HuggingFace: soniqo/Parakeet-TDT-v3-ONNX

Pag - aasikaso

Masalimuot Halaga
Wika114
Pagiging KuwalenteINT8
DecoderMABUTING Kasakiman
Real-time factorRTF 0.12 (Tiper150 m para sa 1.5 s audio)
Bilis ng hardwareNNAPI (Android), QNN Hexagon DSP (Linux)

Kotlin (Android)

val modelDir = ModelManager.ensureModels(context)
val pipeline = SpeechPipeline(SpeechConfig(modelDir = modelDir))
pipeline.events.collect { event ->
    when (event) {
        is SpeechEvent.TranscriptionCompleted -> println(event.text)
        else -> {}
    }
}
pipeline.start()
pipeline.pushAudio(samples) // 16kHz mono float32

CAPI (Linux)

speech_config_t cfg = speech_config_default();
cfg.model_dir = "/opt/models";
speech_pipeline_t p = speech_create(cfg, on_event, NULL);
speech_start(p);
speech_push_audio(p, samples, 512);

Pinagmulan na kodigo: github.com/soniqo/speech-android