Waqf-AI

egyptian-arabic-eot-qwen35-0.8b

README

License: apache-2.0

The problem

Voice activity detection can detect silence, but silence alone does not reveal whether someone has finished speaking. In Egyptian Arabic, a speaker may pause while they are still dictating a number, revising a phrase, or about to complete a request. An EOT detector adds the semantic decision:

EOT — the turn is complete; the agent may answer.
ONGOING — the user is still speaking; the agent should wait.

This checkpoint is intended as a text-based semantic signal alongside an audio VAD, not as a replacement for VAD.

Training data

Fine-tuned on Waqf-AI/egyptian-arabic-turn-detection:

Table with columns: Split, Rows, Use
Split	Rows	Use
train	21,280	fine-tuning
validation	1,200	reserved
test	2,516	held-out evaluation

The dataset is Egyptian Arabic (arz), context-aware, and split by dialogue_id so truncated variants of the same utterance do not leak between training and test. It is synthetic and author-labeled; results therefore measure performance on this task distribution, not a guarantee of real-call performance.

Method

Base model: Qwen/Qwen3.5-0.8B.

Two decision tokens were added and learned with causal-LM supervision:

<|turn_done|> — EOT
<|turn_continue|> — ONGOING

At inference, the score is the normalized probability of <|turn_done|> versus those two decision tokens. Training used full fine-tuning for 3 epochs, learning rate 2e-5, batch size 4, gradient accumulation 8, and a maximum sequence length of 512.

Held-out results

Evaluated once on the dataset's grouped test split (2,516 rows; 1,267 EOT, 1,249 ONGOING):

Table with columns: Model, ROC-AUC, Best balanced accuracy, EOT recall, ONGOING specificity
Model	ROC-AUC	Best balanced accuracy	EOT recall	ONGOING specificity
Qwen3.5-0.8B base	0.500	50.0%	100.0%	0.0%
This fine-tuned checkpoint	0.955	90.6%	92.3%	88.9%

The selected equal-cost operating threshold is 0.51. In a voice agent, false EOT causes an interruption and is usually more costly than a late response; tune the threshold on your own validation calls accordingly.

Examples

With preceding agent context في اي خدمه (“How can I help?”):

Table with columns: Transcript so far, Expected decision, Why
Transcript so far	Expected decision	Why
`تعالي امم انا`	ONGOING	incomplete predicate
`تعالي امم انا كنت عايز`	ONGOING	more information is required
`تعالي امم انا كنت عايز اسال عن الفاتوره`	EOT	complete request
`رقمي صفر واحد صفر`	ONGOING	likely mid-number dictation

Minimal inference

python
import torch
from transformers import AutoModelForMultimodalLM, AutoTokenizer

MODEL_ID = "Waqf-AI/egyptian-arabic-eot-qwen35-0.8b"
DONE, CONTINUE = "<|turn_done|>", "<|turn_continue|>"

tokenizer = AutoTokenizer.from_pretrained(MODEL_ID, padding_side="left")
model = AutoModelForMultimodalLM.from_pretrained(
    MODEL_ID, dtype=torch.bfloat16, device_map="cuda"
).eval()

context = "agent: في اي خدمه"
text = "تعالي امم انا كنت عايز اسال عن الفاتوره"
prompt = (
    "حدد هل المتحدث أنهى دوره أم سيكمل الكلام. أجب برمز واحد فقط: "
    f"{DONE} أو {CONTINUE}.\n\nسياق المحادثة:\n{context}"
    f"\n\nكلام المستخدم حتى الآن:\n{text}\nالقرار:"
)
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
with torch.inference_mode():
    logits = model(**inputs).logits[:, -1, :]
ids = tokenizer.convert_tokens_to_ids([CONTINUE, DONE])
p_eot = torch.softmax(logits[:, ids].float(), dim=-1)[0, 1].item()
print(f"p(EOT) = {p_eot:.3f}")

Limitations and responsible use

This is a text-only model. It does not hear prosody, interruptions, or non-speech audio.
It was trained on synthetic, author-labeled Egyptian Arabic examples. Validate on consented, representative production transcripts before deployment.
Use a conservative threshold and monitor false EOT decisions: an interruption is typically more harmful than a small response delay.
It is not intended for high-stakes decisions or as the sole control for a safety-critical system.

Available on FriendliAI

Dedicated Endpoints

Run this model inference on single tenant GPU with unmatched speed and reliability at scale.

Learn more

Model Details

Model Provider

Waqf-AI

Model Tree

Base

Qwen/Qwen3.5-0.8B

Fine-tuned

this model

Input Modalities

TextImageVideo

Output Modalities

Text

Supported Functionality