Language model and beam search results, plus two more failure log entries

README.md

@@ -71,6 +71,63 @@ agglutinative language, but if you are comparing against word-rate figures
 published elsewhere, that is the number. No open model we tested is close to the
 15% WER that would make Amharic recognition feel solved.
 
+## Does an Amharic language model help?
+
+A CTC model decides each audio frame independently and has no idea what an
+Amharic word is, so when the audio is ambiguous it emits letters that spell
+nothing. An n-gram language model rescores candidates during beam search and
+pulls the output toward real words.
+
+We trained a 5-gram on **12,868,562 lines** of open Amharic text,
+with every test and validation sentence removed. Alpha and beta were tuned on
+the **validation** split and applied unchanged to test.
+
+| model | WER greedy | WER + LM | change | CER greedy | CER + LM | change |
+|---|---|---|---|---|---|---|
+| `Ethio-ASR-amharic` | 0.2845 | **0.2483** | **-12.7%** | 0.0946 | 0.0965 | +2.0% |
+| `Ethio-ASR-multilingual-600M` | 0.2984 | **0.2566** | **-14.0%** | 0.0991 | 0.1034 | +4.3% |
+| `Ethio-ASR-multilingual-1B` | 0.3894 | **0.3154** | **-19.0%** | 0.1308 | 0.1370 | +4.8% |
+
+**Word error rate improves 13 to 19% relative. Character error rate gets
+slightly worse.** That is not a contradiction, it is the language model working
+as designed. It pulls output toward real words, so whole words match far more
+often. When it settles on the wrong real word it changes several characters at
+once, so character error ticks up slightly. For anything a person reads, the
+word rate is what matters.
+
+**The best result on this benchmark is now 24.83% WER**, from
+`Ethio-ASR-amharic` with the language model, against 28.45% without it.
+
+Two caveats. The best alpha and beta both landed on the **edge of the search
+grid**, so a wider search would probably find a better setting. And the trained
+model is 17 GB unpruned, which is impractical to distribute; a pruned version
+is the obvious next step.
+
+### Decontamination
+
+Removing test sentences from 12.8 million lines of text is where a language
+model quietly cheats. Exact whole-line matching found **zero** contaminating
+lines. Substring matching over all 2,990 held-out sentences found **four**,
+buried inside longer paragraphs. Small, but they were there, and nothing would
+have reported an error if we had missed them.
+
+## Does beam search change the ranking?
+
+Greedy decoding is not neutral between model families, so this checks whether it
+disadvantaged the sequence-to-sequence models.
+
+| model | beam width 5 |
+|---|---|
+| `Ethio-ASR-amharic` | not yet rerun, see FAILURES |
+| `Ethio-ASR-multilingual-600M` | not yet rerun, see FAILURES |
+| `Ethio-ASR-multilingual-1B` | not yet rerun, see FAILURES |
+| `shook-medium-amharic-2k` | 0.1147 to 0.1127 CER, 0.2943 to 0.2894 WER |
+| `shook-tiny-amharic-stage2-polish` | 0.2735 to 0.2453 CER, 0.6200 to 0.5812 WER |
+
+It does not change the ranking. Whisper gains a little and pays heavily in
+speed, from 714 seconds to 4,151 for the medium model. The CTC runs hit a bug
+and have not been rerun.
+
 ## Speed
 
 Measured on the same run, one A100, batch size 16. This is throughput for bulk

docs/FAILURES.md

@@ -251,3 +251,47 @@ real at p = 0.001. Reading marginal intervals would have given the wrong answer.
 6. **Medians, when your data has humans in it.**
 
 None of this made the benchmark better. It made it *true*, which took longer.
+
+---
+
+## 11. The same bug, in two files, in one night
+
+`pyctcdecode` requires every label in the alphabet to be unique. Our vocabulary
+mapping sent four special tokens (`<pad>`, `<s>`, `</s>`, `<unk>`) to the empty
+string, because the CTC blank is conventionally empty and the rest looked like
+they should be too.
+
+```
+ValueError: Alphabet contains duplicate entries, this is not allowed.
+```
+
+It killed the language model phase. We fixed it, redeployed, and the phase ran.
+
+Then the beam search phase failed with the identical error, because it had its
+own copy of the same vocabulary code and only one copy had been fixed.
+
+This is the second time in this project that fixing a bug in one file left the
+same bug live in another. The first was a decode limit that truncated a model,
+fixed in one script and retyped into the next two days later.
+
+> A bug fixed in one place is not fixed. Either the code is shared, or the check
+> that catches it runs over everything.
+
+## 12. A broken baseline hiding inside a working result
+
+The language model phase reported its own greedy baseline alongside the language
+model result, so the improvement could be read off directly. Those greedy numbers
+came out around **3.5 CER**, which is impossible.
+
+The cause was the fix above. De-duplicating the alphabet appends a suffix to
+repeated labels, and the phase's hand-rolled greedy decoder emitted those
+suffixes as literal text. The language model path was unaffected because
+`pyctcdecode` handles the alphabet properly.
+
+So the headline number was correct and the comparison beside it was nonsense.
+Anyone reading "3.55 improves to 0.10" would have concluded the language model
+delivers a 97% improvement. The real figure, measured against the verified
+benchmark run, is 13 to 19% on word error rate.
+
+> Never compute a baseline twice. Compare against the number you already
+> measured and trust, not against a fresh reimplementation of it.

results/language_model.json

@@ -0,0 +1,64 @@
+{
+  "language_model": {
+    "corpus_lines": 12868562,
+    "held_out_lines_dropped": 4,
+    "order": 5,
+    "alpha": 0.3,
+    "beta": 1.5,
+    "note": "alpha and beta tuned on the validation split, applied unchanged to test",
+    "models": {
+      "badrex/Ethio-ASR-multilingual-600M": {
+        "cer_greedy": 0.0990624900009599,
+        "cer_lm": 0.10336927654945126,
+        "wer_greedy": 0.2983765690376569,
+        "wer_lm": 0.25660251046025107,
+        "cer_rel_change": 0.04347545219638255,
+        "wer_rel_change": -0.1400044873233116
+      },
+      "badrex/Ethio-ASR-multilingual-1B": {
+        "cer_greedy": 0.1307842447125076,
+        "cer_lm": 0.1370236457300099,
+        "wer_greedy": 0.3893891213389121,
+        "wer_lm": 0.3153807531380753,
+        "cer_rel_change": 0.047707589176493544,
+        "wer_rel_change": -0.19006275251439875
+      },
+      "badrex/Ethio-ASR-amharic": {
+        "cer_greedy": 0.0946085175823121,
+        "cer_lm": 0.09648993696605127,
+        "wer_greedy": 0.28451882845188287,
+        "wer_lm": 0.2483347280334728,
+        "cer_rel_change": 0.019886363636363633,
+        "wer_rel_change": -0.12717647058823534
+      }
+    }
+  },
+  "beam_search": {
+    "beams": 5,
+    "models": {
+      "badrex/Ethio-ASR-amharic": {
+        "error": "alphabet de-duplication bug, not yet rerun"
+      },
+      "badrex/Ethio-ASR-multilingual-600M": {
+        "error": "alphabet de-duplication bug, not yet rerun"
+      },
+      "badrex/Ethio-ASR-multilingual-1B": {
+        "error": "alphabet de-duplication bug, not yet rerun"
+      },
+      "b1n1yam/shook-medium-amharic-2k": {
+        "cer_greedy": 0.11470898793715803,
+        "cer_beam": 0.11266118452628547,
+        "wer_greedy": 0.29425941422594143,
+        "wer_beam": 0.28940585774058575,
+        "seconds": 4151.0
+      },
+      "b1n1yam/shook-tiny-amharic-stage2-polish": {
+        "cer_greedy": 0.2735097430646658,
+        "cer_beam": 0.24526925415160145,
+        "wer_greedy": 0.6200167364016737,
+        "wer_beam": 0.5811882845188284,
+        "seconds": 614.9
+      }
+    }
+  }
+}