ModelÜberprüftExtern

Parakeet Realtime Eou 120m Coreml

nvidia-open-model-license

Quellehuggingface.co/aoiandroid/parakeet-realtime-eou-120m-coreml/tree/1b8ea0eb7770cd61046070cc7500ea4901e16541

ArtOtherBasismodellnvidia/parakeet_realtime_eou_120m-v1VersionRevision 1b8ea0eVerleger@aoiandroidCBewertung

Modellquelle

Art: Other
Dateigröße: 243 B–203 MB
Basismodell: nvidia/parakeet_realtime_eou_120m-v1
Version: vRevision 1b8ea0e
Aktualisiert: 5. Okt. 2026
Verleger: @aoiandroid
Downloads: 11
Parameter: 120M
Quelle: Hugging Face

Quellenbeschreibung

Parakeet Realtime EOU 120M — CoreML

CoreML conversion of [nvidia/parakeet-realtime-eou-120m-v1](https://huggingface.co/nvidia/parakee t-realtime-eou-120m-v1) for streaming speech recognition with end-of-utterance detection on Apple Silicon.

Used by FluidAudio for real-time transcription.

Models

The RNNT pipeline is split into three CoreML models, exported at two chunk sizes:

Model	Description
`streaming_encoder.mlmodelc`	FastConformer encoder with loopback state caching
`decoder.mlmodelc`	1-layer LSTM decoder (640 hidden units)
`joint_decision.mlmodelc`	Joint network for token prediction + EOU detection

Chunk Size Variants

Variant	Latency	WER (test-clean)	RTFx (M2)
`160ms/`	160ms	8.29%	4.78x
`320ms/`	320ms	4.87%	12.48x

Benchmarked on LibriSpeech test-clean (2620 files, 5.40h audio) on Apple M2.

Usage with FluidAudio

import FluidAudio

let manager = StreamingEouAsrManager()
await manager.initialize()

// Transcribe with EOU detection
await manager.startStreaming(
    eouCallback: { transcript in
        print("Utterance complete: \(transcript)")
    },
    partialCallback: { partial in
        print("Partial: \(partial)")
    }
)

// Feed audio chunks as they arrive
await manager.feedAudio(samples)

CLI

Transcribe a file

swift run fluidaudio parakeet-eou --input audio.wav

Benchmark

swift run -c release fluidaudio parakeet-eou --benchmark --chunk-size 320

Architecture

120M parameter RNNT (Recurrent Neural Network Transducer) with:

Encoder: 17-layer FastConformer with cache-aware streaming
Decoder: 1-layer LSTM, 640 hidden size
Joint: Linear projection with 1027 output classes (1024 tokens + EOU token + SOS + blank)
EOU token: ID 1024 signals end-of-utterance

Streaming State

The encoder maintains loopback state between chunks: ┌─────────────────────┬──────────────────┬───────────────────────┐ │ State │ Shape │ Description │ ├─────────────────────┼──────────────────┼───────────────────────┤ │ preCache │ [1, 128, N] │ Mel-level context │ ├─────────────────────┼──────────────────┼───────────────────────┤ │ cacheLastChannel │ [17, 1, 70, 512] │ Conformer layer cache │ ├─────────────────────┼──────────────────┼───────────────────────┤ │ cacheLastTime │ [17, 1, 512, 8] │ Temporal cache │ ├─────────────────────┼──────────────────┼───────────────────────┤ │ cacheLastChannelLen │ [1] │ Cache length tracking │ └─────────────────────┴──────────────────┴───────────────────────┘ Export

Converted from PyTorch using coremltools. To re-export:

python3 Scripts/ParakeetEOU/Conversion/convert_split_encoder.py
--output-dir Models/ParakeetEOU
--model-id nvidia/parakeet-realtime-eou-120m-v1

License

NVIDIA Open Model License — see https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/.

Original model: https://huggingface.co/nvidia/parakeet-realtime-eou-120m-v1

Quellen

1 Quelle

Hugging FacePrimärVersion 1b8ea0eb7770cd61046070cc7500ea4901e16541

huggingface.co/aoiandroid/parakeet-realtime-eou-120m-coreml/tree/1b8ea0eb7770cd61046070cc7500ea4901e16541

Verifiziert 19. Aug.

Modellartefakte

41 Artefakte

1280ms/decoder.mlmodelc/analytics/coremldata.bin

bin · 243 B · SHA-256 3996975a8cbc…2b49 · Hugging Face

Herunterladen

1280ms/decoder.mlmodelc/coremldata.bin

bin · 497 B · SHA-256 c3ccbff963d8…dc34 · Hugging Face

Herunterladen

1280ms/decoder.mlmodelc/weights/weight.bin

bin · 7,51 MB · SHA-256 0b4cacecdcd9…b2f4 · Hugging Face

Herunterladen

1280ms/joint_decision.mlmodelc/analytics/coremldata.bin

bin · 243 B · SHA-256 5bca32ad130d…55b5 · Hugging Face

Herunterladen

1280ms/joint_decision.mlmodelc/coremldata.bin

bin · 493 B · SHA-256 22d4abc4625b…9625 · Hugging Face

Herunterladen

1280ms/joint_decision.mlmodelc/weights/weight.bin

bin · 2,66 MB · SHA-256 7039b2010a26…84b4 · Hugging Face

Herunterladen

1280ms/parakeet_eou_preprocessor.mlmodelc/analytics/coremldata.bin

bin · 243 B · SHA-256 f4ada8b0b99a…ad63 · Hugging Face

Herunterladen

1280ms/parakeet_eou_preprocessor.mlmodelc/coremldata.bin

bin · 422 B · SHA-256 dc7252fa4762…ec5e · Hugging Face

Herunterladen

1280ms/parakeet_eou_preprocessor.mlmodelc/weights/weight.bin

bin · 1,13 MB · SHA-256 009bba4fde82…291d · Hugging Face

Herunterladen

1280ms/streaming_encoder.mlmodelc/analytics/coremldata.bin

bin · 243 B · SHA-256 d0a3c84022a9…4004 · Hugging Face

Herunterladen

1280ms/streaming_encoder.mlmodelc/coremldata.bin

bin · 671 B · SHA-256 41ce3f96c3d6…bb78 · Hugging Face

Herunterladen

1280ms/streaming_encoder.mlmodelc/weights/weight.bin

bin · 203 MB · SHA-256 6c71acb590ce…4b17 · Hugging Face

Herunterladen

1280ms/streaming_encoder.mlpackage/Data/com.apple.CoreML/weights/weight.bin

bin · 203 MB · SHA-256 6c71acb590ce…4b17 · Hugging Face

Herunterladen

160ms/decoder.mlmodelc/analytics/coremldata.bin

bin · 243 B · SHA-256 3996975a8cbc…2b49 · Hugging Face

Herunterladen

160ms/decoder.mlmodelc/coremldata.bin

bin · 497 B · SHA-256 c3ccbff963d8…dc34 · Hugging Face

Herunterladen

160ms/decoder.mlmodelc/weights/weight.bin

bin · 7,51 MB · SHA-256 0b4cacecdcd9…b2f4 · Hugging Face

Herunterladen

160ms/decoder.mlpackage/Data/com.apple.CoreML/weights/weight.bin

bin · 7,51 MB · SHA-256 0b4cacecdcd9…b2f4 · Hugging Face

Herunterladen

160ms/joint_decision.mlmodelc/analytics/coremldata.bin

bin · 243 B · SHA-256 5bca32ad130d…55b5 · Hugging Face

Herunterladen

160ms/joint_decision.mlmodelc/coremldata.bin

bin · 493 B · SHA-256 22d4abc4625b…9625 · Hugging Face

Herunterladen

160ms/joint_decision.mlmodelc/weights/weight.bin

bin · 2,66 MB · SHA-256 7039b2010a26…84b4 · Hugging Face

Herunterladen

160ms/joint_decision.mlpackage/Data/com.apple.CoreML/weights/weight.bin

bin · 2,66 MB · SHA-256 7039b2010a26…84b4 · Hugging Face

Herunterladen

160ms/parakeet_eou_preprocessor.mlmodelc/analytics/coremldata.bin

bin · 243 B · SHA-256 f4ada8b0b99a…ad63 · Hugging Face

Herunterladen

160ms/parakeet_eou_preprocessor.mlmodelc/coremldata.bin

bin · 422 B · SHA-256 dc7252fa4762…ec5e · Hugging Face

Herunterladen

160ms/parakeet_eou_preprocessor.mlmodelc/weights/weight.bin

bin · 1,13 MB · SHA-256 009bba4fde82…291d · Hugging Face

Herunterladen

160ms/parakeet_eou_preprocessor.mlpackage/Data/com.apple.CoreML/weights/weight.bin

bin · 579 KB · SHA-256 f257ad1ac115…8e53 · Hugging Face

Herunterladen

160ms/streaming_encoder.mlmodelc/analytics/coremldata.bin

bin · 243 B · SHA-256 a981b257db79…0356 · Hugging Face

Herunterladen

160ms/streaming_encoder.mlmodelc/coremldata.bin

bin · 670 B · SHA-256 e762abc60d99…037a · Hugging Face

Herunterladen

160ms/streaming_encoder.mlmodelc/weights/weight.bin

bin · 203 MB · SHA-256 12cd781a4300…070c · Hugging Face

Herunterladen

160ms/streaming_encoder.mlpackage/Data/com.apple.CoreML/weights/weight.bin

bin · 203 MB · SHA-256 12cd781a4300…070c · Hugging Face

Herunterladen

320ms/decoder.mlmodelc/analytics/coremldata.bin

bin · 243 B · SHA-256 3996975a8cbc…2b49 · Hugging Face

Herunterladen

320ms/decoder.mlmodelc/coremldata.bin

bin · 497 B · SHA-256 c3ccbff963d8…dc34 · Hugging Face

Herunterladen

320ms/decoder.mlmodelc/weights/weight.bin

bin · 7,51 MB · SHA-256 0b4cacecdcd9…b2f4 · Hugging Face

Herunterladen

320ms/decoder.mlpackage/Data/com.apple.CoreML/weights/weight.bin

bin · 7,51 MB · SHA-256 0b4cacecdcd9…b2f4 · Hugging Face

Herunterladen

320ms/joint_decision.mlmodelc/analytics/coremldata.bin

bin · 243 B · SHA-256 5bca32ad130d…55b5 · Hugging Face

Herunterladen

320ms/joint_decision.mlmodelc/coremldata.bin

bin · 493 B · SHA-256 22d4abc4625b…9625 · Hugging Face

Herunterladen

320ms/joint_decision.mlmodelc/weights/weight.bin

bin · 2,66 MB · SHA-256 7039b2010a26…84b4 · Hugging Face

Herunterladen

320ms/joint_decision.mlpackage/Data/com.apple.CoreML/weights/weight.bin

bin · 2,66 MB · SHA-256 7039b2010a26…84b4 · Hugging Face

Herunterladen

320ms/streaming_encoder.mlmodelc/analytics/coremldata.bin

bin · 243 B · SHA-256 ae0a559f8d53…aec5 · Hugging Face

Herunterladen

320ms/streaming_encoder.mlmodelc/coremldata.bin

bin · 670 B · SHA-256 b57dfe32ed2a…43c3 · Hugging Face

Herunterladen

320ms/streaming_encoder.mlmodelc/weights/weight.bin

bin · 203 MB · SHA-256 ff553fcb4f45…6822 · Hugging Face

Herunterladen

320ms/streaming_encoder.mlpackage/Data/com.apple.CoreML/weights/weight.bin

bin · 203 MB · SHA-256 ff553fcb4f45…6822 · Hugging Face

Herunterladen

Quellenauszüge

3 Auszüge

Readme11. Julihuggingface.co/aoiandroid/parakeet-realtime-eou-120m-coreml

--- license: other license_name: nvidia-open-model-license license_link: >- https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/ language: - en metrics: - wer library_name: nemo tags: - speech-recognition - FastConformer - end-of-utterance - voice agent pipeline_tag: automatic-speech-recognition base_model: - nvidia/parakeet_realtime_eou_120m-v1 base_model_relation: finetune --- # Parakeet Realtime EOU 120M — CoreML CoreML conversion of [nvidia/parakeet-realtime-eou-120m-v1](https://huggingface.co/nvidia/parakee t-realtime-eou-120m-v1) for streaming speech recognition with end-of-utterance detection on Apple Silicon. Used by [FluidAudio](https://github.com/FluidInference/FluidAudio) for real-time transcription. ## Models The RNNT pipeline is split into three CoreML models, exported at two chunk sizes: | Model | Description | |-------|-------------| | `streaming_encoder.mlmodelc` | FastConformer encoder with loopback state caching | | `decoder.mlmodelc` | 1-layer LSTM decoder (640 hidden units) | | `joint_decision.mlmodelc` | Joint network for token prediction + EOU detection | ### Chunk Size Variants | Variant | Latency | WER (test-clean) | RTFx (M2) | |---------|---------|-------------------|------------| | `160ms/` | 160ms | 8.29% | 4.78x | | `320ms/` | 320ms | 4.87% | 12.48x | Benchmarked on LibriSpeech test-clean (2620 files, 5.40h audio) on Apple M2. ## Usage with FluidAudio ```swift import FluidAudio let manager = StreamingEouAsrManager() await manager.initialize() // Transcribe with EOU detection await manager.startStreaming( eouCallback: { transcript in print("Utterance complete: \(transcript)") }, partialCallback: { partial in print("Partial: \(partial)") } ) // Feed audio chunks as they arrive await manager.feedAudio(samples) ``` CLI # Transcribe a file swift run fluidaudio parakeet-eou --input audio.wav # Benchmark swift run -c release fluidaudio parakeet-eou --benchmark --chunk-size 320 Architecture 120M parameter RNNT (Recurrent Neural Network Transducer) with: - Encoder: 17-layer FastConformer with cache-aware streaming - Decoder: 1-layer LSTM, 640 hidden size - Joint: Linear projection with 1027 output classes (1024 tokens + EOU token + SOS + blank) - EOU token: ID 1024 signals end-of-utterance Streaming State The encoder maintains loopback state between chunks: ┌───────────────...

Note19. Aug.huggingface.co/aoiandroid/parakeet-realtime-eou-120m-coreml

Source context: 11 downloads · 0 likes · Pipeline automatic-speech-recognition · Library nemo · Repo aoiandroid/parakeet-realtime-eou-120m-coreml

Note17. Julihuggingface.co/aoiandroid/parakeet-realtime-eou-120m-coreml

Source context: 24 downloads · 0 likes · Pipeline automatic-speech-recognition · Library nemo · Repo aoiandroid/parakeet-realtime-eou-120m-coreml