Source code
Revision control
Copy as Markdown
Other Tools
# Speech recognition model manifest.
#
# Ordered list of models that can perform speech recognition. The first model
# matching a BCP-47 tag is selected.
#
# Fields
# ------
# id : short identifier used in generated symbol names and in the pref
# media.webspeech.recognition.model.<locale_prefix>
# supported_locales : Locale tags this artifact recognizes, spelled as the model
# expects them; SpeechModelFor() negotiates a request against
# them, returning the model id the utility process passes around
# plus that locale (SpeechRecognitionModelMapping.h).
# streaming : true = cache-aware streaming RNN-T/transducer (parakeet_capi_stream_*)
# false = offline CTC batch path (parakeet_capi_transcribe_pcm)
# repo : ModelHub repository (org/name)
# filename : GGUF file name inside the repository
# revision : ModelHub revision tag
# size_mb : Approximate download size shown in the permission prompt (integer MB)
# quant : Quantization identifier (informational)
# latency_ms : Approximate per-chunk inference latency in ms; 0 = not yet measured
# display_name : Human-readable model name shown in about:addons' "Manage
# On-Device AI Models" panel. Several distinct upstream models are
# rehosted under the single "speech-recognition/parakeet" ModelHub
# bucket (repo), so that bucket's path cannot be used to derive a
# precise name; this is the real per-artifact name instead.
# hf_url : Upstream Hugging Face model card for this artifact, shown as the
# "View on Hugging Face" link. Not derivable from repo/revision
# (see display_name) since the ModelHub bucket path and the real
# upstream Hugging Face repo path differ for this feature.
models:
# --- English streaming model (EOU-aware, RNN-T transducer) ---
- id: english
supported_locales:
- en
streaming: true
repo: "speech-recognition/parakeet"
filename: "realtime_eou_120m-v1-q5_k.gguf"
revision: "main"
size_mb: 141
quant: "q5_k"
latency_ms: 0
display_name: "Parakeet Realtime EOU 120M v1"
# --- English TDT-CTC 110m (offline CTC; multiple quants) ---
- id: english_tdt_ctc_q4_k
supported_locales:
- en
streaming: false
repo: "speech-recognition/parakeet"
filename: "tdt_ctc-110m-q4_k.gguf"
revision: "main"
size_mb: 131
quant: "q4_k"
latency_ms: 0
display_name: "Parakeet TDT-CTC 110M"
- id: english_tdt_ctc_q5_k
supported_locales:
- en
streaming: false
repo: "speech-recognition/parakeet"
filename: "tdt_ctc-110m-q5_k.gguf"
revision: "main"
size_mb: 143
quant: "q5_k"
latency_ms: 0
display_name: "Parakeet TDT-CTC 110M"
- id: english_tdt_ctc_q6_k
supported_locales:
- en
streaming: false
repo: "speech-recognition/parakeet"
filename: "tdt_ctc-110m-q6_k.gguf"
revision: "main"
size_mb: 156
quant: "q6_k"
latency_ms: 0
display_name: "Parakeet TDT-CTC 110M"
- id: english_tdt_ctc_q8_0
supported_locales:
- en
streaming: false
repo: "speech-recognition/parakeet"
filename: "tdt_ctc-110m-q8_0.gguf"
revision: "main"
size_mb: 178
quant: "q8_0"
latency_ms: 0
display_name: "Parakeet TDT-CTC 110M"
- id: english_tdt_ctc_f16
supported_locales:
- en
streaming: false
repo: "speech-recognition/parakeet"
filename: "tdt_ctc-110m-f16.gguf"
revision: "main"
size_mb: 268
quant: "f16"
latency_ms: 0
display_name: "Parakeet TDT-CTC 110M"
# --- Multilingual (nemotron, streaming RNN-T) ---
# Last, so English goes to the English models above even though this one
# also lists en-US/en-GB.
- id: multilingual
# Straight from the model's own prompt dictionary
# (parakeet.prompt.dictionary.keys), minus its malformed keys (enGB, esES) and
# its "auto" default, which is what passing no locale at all selects.
# The model card claims quality for a subset, in decreasing order:
# core: en-US en-GB es-US es-ES fr-FR fr-CA it-IT pt-BR
# pt-PT nl-NL de-DE tr-TR ru-RU ar-AR hi-IN ja-JP
# ko-KR vi-VN uk-UA
# broad coverage: pl-PL sv-SE cs-CZ nb-NO da-DK bg-BG fi-FI hr-HR
# sk-SK zh-CN hu-HU ro-RO et-EE
# adaptation-ready: el-GR lt-LT lv-LV mt-MT sl-SI he-IL th-TH nn-NO
# the rest the model accepts but the card does not claim.
supported_locales:
- en-US
- en
- en-GB
- es-ES
- es-US
- es
- zh-CN
- zh-ZH
- zh-TW
- hi-IN
- hi
- hi-HI
- ar-AR
- ar
- fr-FR
- fr
- de-DE
- de
- ja-JP
- ja-JA
- ru-RU
- ru
- pt-BR
- pt-PT
- pt
- ko-KR
- ko
- ko-KO
- it-IT
- it
- nl-NL
- nl
- pl-PL
- pl
- tr-TR
- tr
- uk-UA
- uk
- ro-RO
- ro
- el-GR
- el
- cs-CZ
- cs
- hu-HU
- hu
- sv-SE
- sv
- da-DK
- da
- fi-FI
- fi
- no-NO
# Quoted: YAML reads a bare no as false.
- "no"
- nb-NO
- nb
- nn-NO
- nn
- sk-SK
- sk
- hr-HR
- hr
- bg-BG
- bg
- lt-LT
- lt
- et-EE
- et
- lv-LV
- lv
- sl-SI
- sl
- th-TH
- vi-VN
- id-ID
- ms-MY
- bn-IN
- ur-PK
- fa-IR
- ta-IN
- te-IN
- mr-IN
- gu-IN
- kn-IN
- ml-IN
- si-LK
- ne-NP
- km-KH
- sw-KE
- am-ET
- ha-NG
- zu-ZA
- yo-NG
- ig-NG
- af-ZA
- rw-RW
- so-SO
- ny-MW
- ln-CD
- or-KE
- he-IL
- ku-TR
- az-AZ
- ka-GE
- hy-AM
- uz-UZ
- tg-TJ
- ky-KG
- qu-PE
- ay-BO
- gn-PY
- nah-MX
- mi-NZ
- haw-US
- sm-WS
- to-TO
- fr-CA
- mt-MT
streaming: true
repo: "speech-recognition/parakeet"
filename: "nemotron-3.5-asr-streaming-0.6b-q5_k.gguf"
revision: "main"
size_mb: 785
quant: "q5_k"
latency_ms: 0
display_name: "Nemotron 3.5 ASR Streaming 0.6B"