Fix model card images with Hub resolve URLs

README.md

@@ -79,7 +79,7 @@ model-index:
 ---
 
 <div align="center">
-  <img src="assets/banner.svg" alt="VaaniCall — multilingual telephony ASR for India" width="100%">
+  <img src="https://huggingface.co/TieIncred/VaaniCall/resolve/main/assets/banner.svg" alt="VaaniCall — multilingual telephony ASR for India" width="100%">
 </div>
 
 <br>
@@ -92,7 +92,7 @@ By [Tausif Iqbal](https://huggingface.co/TieIncred). Built on NVIDIA NeMo FastCo
 VaaniCall transcribes a phone call without a language code. It infers language and script, and it is trained through a 21-stage telephony pipeline so AMR, packet loss, and handset noise look like training data — not a domain shift.
 
 <div align="center">
-  <img src="assets/languages.svg" alt="Supported languages: Hindi, Bengali, Tamil, Telugu, Kannada, Malayalam, Marathi, Gujarati, Punjabi, Odia, English" width="100%">
+  <img src="https://huggingface.co/TieIncred/VaaniCall/resolve/main/assets/languages.svg" alt="Supported languages: Hindi, Bengali, Tamil, Telugu, Kannada, Malayalam, Marathi, Gujarati, Punjabi, Odia, English" width="100%">
 </div>
 
 ---
@@ -114,7 +114,7 @@ VaaniCall transcribes a phone call without a language code. It infers language a
 All numbers are **Word Error Rate** (lower is better). Evaluations use **500 clips per language**.
 
 <div align="center">
-  <img src="assets/wer_overall_vs_whisper.png" alt="Overall WER: VaaniCall vs Whisper large-v3 on IndicVoices and FLEURS" width="100%">
+  <img src="https://huggingface.co/TieIncred/VaaniCall/resolve/main/assets/wer_overall_vs_whisper.png" alt="Overall WER: VaaniCall vs Whisper large-v3 on IndicVoices and FLEURS" width="100%">
 </div>
 
 | Benchmark | Condition | VaaniCall | Whisper large-v3 | Relative WER cut |
@@ -126,7 +126,7 @@ All numbers are **Word Error Rate** (lower is better). Evaluations use **500 cli
 Whisper large-v3 still leads on **English**. VaaniCall is the Indic-language model.
 
 <div align="center">
-  <img src="assets/telephony_robustness.png" alt="VaaniCall WER stays flat from clean to telephonic audio" width="100%">
+  <img src="https://huggingface.co/TieIncred/VaaniCall/resolve/main/assets/telephony_robustness.png" alt="VaaniCall WER stays flat from clean to telephonic audio" width="100%">
 </div>
 
 ---
@@ -136,7 +136,7 @@ Whisper large-v3 still leads on **English**. VaaniCall is the Indic-language mod
 ### vs Whisper large-v3 — IndicVoices, telephonic
 
 <div align="center">
-  <img src="assets/wer_indicvoices_telephonic.png" alt="Per-language telephonic WER vs Whisper large-v3" width="100%">
+  <img src="https://huggingface.co/TieIncred/VaaniCall/resolve/main/assets/wer_indicvoices_telephonic.png" alt="Per-language telephonic WER vs Whisper large-v3" width="100%">
 </div>
 
 <details>
@@ -184,7 +184,7 @@ Whisper large-v3 still leads on **English**. VaaniCall is the Indic-language mod
 Held-out read speech. Odia is not in FLEURS.
 
 <div align="center">
-  <img src="assets/wer_fleurs.png" alt="FLEURS out-of-domain WER vs Whisper large-v3" width="100%">
+  <img src="https://huggingface.co/TieIncred/VaaniCall/resolve/main/assets/wer_fleurs.png" alt="FLEURS out-of-domain WER vs Whisper large-v3" width="100%">
 </div>
 
 <details>
@@ -211,7 +211,7 @@ Held-out read speech. Odia is not in FLEURS.
 ### Per-language profile
 
 <div align="center">
-  <img src="assets/wer_by_language.png" alt="VaaniCall WER by language on telephonic IndicVoices" width="100%">
+  <img src="https://huggingface.co/TieIncred/VaaniCall/resolve/main/assets/wer_by_language.png" alt="VaaniCall WER by language on telephonic IndicVoices" width="100%">
 </div>
 
 Strongest Indic languages on phone audio: **Punjabi 14.5% · Bengali 15.4% · Hindi 16.0% · Marathi 17.1%**.
@@ -221,7 +221,7 @@ Strongest Indic languages on phone audio: **Punjabi 14.5% · Bengali 15.4% · Hi
 VaaniCall is a telephony specialist on top of VAANI. On phone-channel audio it improves **9 of 10** Indic languages. On clean audio the two models are statistically tied (24.37% vs 24.38% WER).
 
 <div align="center">
-  <img src="assets/delta_vs_vaani_base.png" alt="WER reduction versus VAANI base under telephonic conditions" width="100%">
+  <img src="https://huggingface.co/TieIncred/VaaniCall/resolve/main/assets/delta_vs_vaani_base.png" alt="WER reduction versus VAANI base under telephonic conditions" width="100%">
 </div>
 
 <details>
@@ -320,7 +320,7 @@ Largest telephonic gaps: Malayalam −5.61 pp, Telugu −5.49 pp. English exclud
 ## How it works
 
 <div align="center">
-  <img src="assets/architecture.svg" alt="Phone audio into FastConformer encoder, RNNT decoder, transcript with no language ID" width="100%">
+  <img src="https://huggingface.co/TieIncred/VaaniCall/resolve/main/assets/architecture.svg" alt="Phone audio into FastConformer encoder, RNNT decoder, transcript with no language ID" width="100%">
 </div>
 
 ### Training
@@ -328,7 +328,7 @@ Largest telephonic gaps: Malayalam −5.61 pp, Telugu −5.49 pp. English exclud
 Two-phase fine-tune from the VAANI FastConformer checkpoint.
 
 <div align="center">
-  <img src="assets/training.svg" alt="Phase 1 language alignment, phase 2 acoustic refinement" width="100%">
+  <img src="https://huggingface.co/TieIncred/VaaniCall/resolve/main/assets/training.svg" alt="Phase 1 language alignment, phase 2 acoustic refinement" width="100%">
 </div>
 
 1. **Language alignment** — encoder frozen; decoder learns script and language mapping.
@@ -343,7 +343,7 @@ Trained on [AI4Bharat IndicVoices](https://huggingface.co/datasets/ai4bharat/Ind
 Every training batch can pass through a 21-stage, GPU-accelerated call-channel simulator.
 
 <div align="center">
-  <img src="assets/pipeline.svg" alt="Talker, handset, codec, network, and line stages of the augmentation pipeline" width="100%">
+  <img src="https://huggingface.co/TieIncred/VaaniCall/resolve/main/assets/pipeline.svg" alt="Talker, handset, codec, network, and line stages of the augmentation pipeline" width="100%">
 </div>
 
 <details>