ModelVérifiéExterne

Dasheng Base Env Encoder

mispeech/dasheng-base fine-tuned (LoRA, merged) to extract environment/background embeddings from speech. Trained on ChristianYang/Env-TTS-Clean with conversation-level AAMSoftmax. Auto-uploaded from training step...

Sourcehuggingface.co/ChristianYang/dasheng-base-env-encoder/tree/e92f375d0ce1f23a736d4b774aab252d522398b7

TypeAudio modelModèle de basemispeech/dasheng-baseVersionRevision e92f375Éditeur@ChristianYangCClasse

Source du modèle

Type: Audio model, Embedding
Taille du fichier: 13,5–326 MB
Modèle de base: mispeech/dasheng-base
Version: vRevision e92f375
Mis à jour: 3 oct. 2026
Éditeur: @ChristianYang
Téléchargements: 3
Tâches: Audio
Source: Hugging Face

Description de la source

dasheng-base-env-encoder

mispeech/dasheng-base fine-tuned (LoRA, merged) to extract environment/background embeddings from speech. Trained on ChristianYang/Env-TTS-Clean with conversation-level AAMSoftmax. Auto-uploaded from training step 100000.

Usage

import torch
from transformers import AutoModel, AutoFeatureExtractor
from huggingface_hub import hf_hub_download

repo = "ChristianYang/dasheng-base-env-encoder"
backbone = AutoModel.from_pretrained(repo, trust_remote_code=True).eval()
fe = AutoFeatureExtractor.from_pretrained(repo, trust_remote_code=True)
head = torch.load(hf_hub_download(repo, "head.pt"), map_location="cpu", weights_only=True)
# pipeline: 16 kHz wav -> fe -> mel -> backbone.encoder tokens [B,T,768]
# -> Conv1d x2 (head["cnn"]) -> masked attentive pooling (head["pool"]) -> 768-d embedding

Head weights (head.pt): two Conv1d(768,768,k=3) layers + attentive pooling (Linear(768,1)). Token count = mel frames // 4.

Sources

1 source

Hugging FacePrincipalVersion e92f375d0ce1f23a736d4b774aab252d522398b7

huggingface.co/ChristianYang/dasheng-base-env-encoder/tree/e92f375d0ce1f23a736d4b774aab252d522398b7

Vérifié 3 oct.

Artefacts du modèle

2 artefacts

head.pt

pt · 13,5 MB · SHA-256 f8c21e9d63a6…2c26 · Hugging Face

Télécharger

model.safetensors

safetensors · 326 MB · SHA-256 830b7b3641e5…4c14 · Hugging Face

Télécharger

Extraits de sources

3 extraits

Readme11 juil.huggingface.co/ChristianYang/dasheng-base-env-encoder

--- license: apache-2.0 base_model: mispeech/dasheng-base --- # dasheng-base-env-encoder `mispeech/dasheng-base` fine-tuned (LoRA, merged) to extract environment/background embeddings from speech. Trained on `ChristianYang/Env-TTS-Clean` with conversation-level AAMSoftmax. Auto-uploaded from training step **100000**. ## Usage ```python import torch from transformers import AutoModel, AutoFeatureExtractor from huggingface_hub import hf_hub_download repo = "ChristianYang/dasheng-base-env-encoder" backbone = AutoModel.from_pretrained(repo, trust_remote_code=True).eval() fe = AutoFeatureExtractor.from_pretrained(repo, trust_remote_code=True) head = torch.load(hf_hub_download(repo, "head.pt"), map_location="cpu", weights_only=True) # pipeline: 16 kHz wav -> fe -> mel -> backbone.encoder tokens [B,T,768] # -> Conv1d x2 (head["cnn"]) -> masked attentive pooling (head["pool"]) -> 768-d embedding ``` Head weights (`head.pt`): two Conv1d(768,768,k=3) layers + attentive pooling (Linear(768,1)). Token count = mel frames // 4.

Note3 oct.huggingface.co/ChristianYang/dasheng-base-env-encoder

Source context: 3 downloads · 0 likes · Repo ChristianYang/dasheng-base-env-encoder

Note17 juil.huggingface.co/ChristianYang/dasheng-base-env-encoder

Source context: 5 downloads · 0 likes · Repo ChristianYang/dasheng-base-env-encoder