Stage verified Inflect v2 release candidate and evaluation evidence

README.md

@@ -15,11 +15,11 @@ tags:
 
 <h1 align="center">Inflect-Micro-v2</h1>
 
-<p align="center"><strong>A complete 24 kHz text-to-waveform TTS system in 9,356,513 parameters.</strong><br>One English voice. CPU-ready. No external vocoder.</p>
-
-<p align="center"><a href="https://huggingface.co/spaces/owensong/Inflect-v2-Private-Playground"><strong>Try Micro and Nano in the browser</strong></a></p>
-
-Inflect-Micro-v2 is designed for local applications where the entire speech stack must fit below **10M parameters**. The count includes text encoding, duration modeling, latent generation, and waveform decoding. It does not hide a separately downloaded vocoder.
+<p align="center"><strong>A complete 24 kHz text-to-waveform TTS system in 9,356,513 parameters.</strong><br>One English voice. CPU-ready. No external vocoder.</p>
+
+<p align="center"><a href="https://huggingface.co/spaces/owensong/Inflect-v2-Private-Playground"><strong>Try Micro and Nano in the browser</strong></a></p>
+
+Inflect-Micro-v2 is designed for local applications where the entire speech stack must fit below **10M parameters**. The count includes text encoding, duration modeling, latent generation, and waveform decoding. It does not hide a separately downloaded vocoder.
 
 ## Listen First
 
@@ -67,9 +67,9 @@ The selected Inflect-Micro-v2 checkpoint scored **4.406 UTMOS22** and **1.23% se
 - All four systems synthesize the same 200 unseen English prompts.
 - Semantic WER uses Whisper-large-v3 and the disclosed English text normalizer.
 - UTMOS22 uses `tarepan/SpeechMOS` v1.2.0 with 5,000 paired bootstrap samples.
-- NISQA dimensions, signal diagnostics, categorized ASR failures, speaker consistency, and multi-seed robustness are under `evaluation/`.
-- `evaluation/asr_robustness.json` separates sentence failures, semantic character errors, tail WER, and category-level errors.
-- Competitor results apply only to the named checkpoints and voices, not every configuration of those projects.
+- NISQA dimensions, signal diagnostics, categorized ASR failures, speaker consistency, and multi-seed robustness are under `evaluation/`.
+- `evaluation/asr_robustness.json` separates sentence failures, semantic character errors, tail WER, and category-level errors.
+- Competitor results apply only to the named checkpoints and voices, not every configuration of those projects.
 
 </details>
 

inflect_nano_v2_frontend.py

@@ -119,6 +119,40 @@ def _digit_words(text: str) -> str:
     return " ".join(_words(int(ch)) for ch in text if ch.isdigit())
 
 
+def _identifier_digits(text: str) -> str:
+    words = []
+    for index, character in enumerate(text):
+        if not character.isdigit():
+            continue
+        words.append("oh" if character == "0" and index > 0 else _words(int(character)))
+    return " ".join(words)
+
+
+def _expand_identifier_token(token: str) -> str:
+    match = re.fullmatch(r"([A-Za-z]?)(\d+)([A-Za-z]?)", token)
+    if match is None:
+        return token
+    prefix, digits, suffix = match.groups()
+    pieces = []
+    if prefix:
+        pieces.append(LETTER_NAMES[prefix.upper()])
+    if len(digits) == 3 or digits.startswith("0"):
+        pieces.append(_identifier_digits(digits))
+    else:
+        pieces.append(_words(int(digits)))
+    if suffix:
+        pieces.append(LETTER_NAMES[suffix.upper()])
+    return " ".join(pieces)
+
+
+def _expand_labeled_identifier(match: re.Match[str]) -> str:
+    return f"{match.group(1)} {_expand_identifier_token(match.group(2))}"
+
+
+def _expand_street_number(match: re.Match[str]) -> str:
+    return _identifier_digits(match.group(1))
+
+
 def _expand_money(match: re.Match[str]) -> str:
     raw = match.group(1).replace(",", "")
     dollars, _, cents = raw.partition(".")
@@ -208,6 +242,18 @@ def normalize_text(text: str) -> str:
         text = re.sub(rf"\b{re.escape(src)}", dst, text, flags=re.IGNORECASE)
 
     text = re.sub(r"\b([A-Z])(?:\.([A-Z]))+\.", lambda m: " ".join(re.findall(r"[A-Z]", m.group(0))), text)
+    text = re.sub(
+        r"\b(apartment|apt\.?|suite|unit|room|flight|extension|order|invoice|locker|aisle|gate)\s+([A-Za-z]?\d{1,4}[A-Za-z]?)\b",
+        _expand_labeled_identifier,
+        text,
+        flags=re.IGNORECASE,
+    )
+    text = re.sub(
+        r"\b(\d{3})(?=\s+(?:North|South|East|West)\b)",
+        _expand_street_number,
+        text,
+        flags=re.IGNORECASE,
+    )
     text = re.sub(r"\$(\d[\d,]*(?:\.\d{1,2})?)", _expand_money, text)
     text = re.sub(r"\b(0?[1-9]|1[0-2])/(0?[1-9]|[12]\d|3[01])/(20\d{2}|19\d{2})\b", _expand_date_slash, text)
     text = re.sub(r"\b(\d{1,2}):(\d{2})\s*([AaPp]\.?\s*[Mm]\.?)?\b", _expand_time, text)

release_manifest.json

@@ -11,20 +11,85 @@
   "standalone_cpu_smoke_audio_seconds": 3.616,
   "files": [
     {
-      "path": "LICENSE",
-      "bytes": 10455,
-      "sha256": "12c48c640d1d0baf158d45175a4238b9227c71c77f50b0d15791b8efc08cd793"
-    },
-    {
-      "path": "THIRD_PARTY_NOTICES.md",
-      "bytes": 892,
-      "sha256": "ef5343cd39713d313aefcfd09bb55fd6606ae2dacabeee13e0abdaa2e5b62589"
+      "path": "assets/matched-evaluation.svg",
+      "bytes": 2513,
+      "sha256": "027f3ad488d19a3d2c5cc86ae4599bd29e50494d4f55377914cd5d954d2f56ca"
     },
     {
       "path": "config.json",
       "bytes": 1992,
       "sha256": "627f66cf339155d4b8524ce78b65230d2e2f33850f6e3f87f2e46c1add7e27d7"
     },
+    {
+      "path": "evaluation/asr.json",
+      "bytes": 862560,
+      "sha256": "756cfa30037a84a36a854adcd44fba2f549b40115d4e68515fab568a79b09cbd"
+    },
+    {
+      "path": "evaluation/asr_failures.json",
+      "bytes": 71601,
+      "sha256": "70e9e6d56db4380eb9a09248faf8b7923b3b96453920864479873c9c1a6547ad"
+    },
+    {
+      "path": "evaluation/asr_robustness.json",
+      "bytes": 34249,
+      "sha256": "624d547740371d4d746a0cb7fbd22e6301c317d83659e7c26e3435ffc9beff46"
+    },
+    {
+      "path": "evaluation/long_form.json",
+      "bytes": 417,
+      "sha256": "eec42c609379720df5426cdbee2d23b96ce2b6308d04edb2d6f8de46ee507ba9"
+    },
+    {
+      "path": "evaluation/long_form_asr.json",
+      "bytes": 12586,
+      "sha256": "353dc41dd4a3f4925a479f1d026a5ac642042c141d6a3335627fc3e5bec48c28"
+    },
+    {
+      "path": "evaluation/long_form_signal.json",
+      "bytes": 14483,
+      "sha256": "ffc932196d48e862ff5b722dbe72068755a72db99bfef91f1e8479e54ae0b3c1"
+    },
+    {
+      "path": "evaluation/multiseed_robustness.json",
+      "bytes": 41769,
+      "sha256": "a577da1dd4632f12d2a728976b02fd18e3aac7a9878385f023242b96ebf96ba9"
+    },
+    {
+      "path": "evaluation/nisqa.json",
+      "bytes": 56921,
+      "sha256": "9c1cbe99dd0994f02da9b30c07e70ca0a0487c2894367ee4c27b8df3d17d8cb9"
+    },
+    {
+      "path": "evaluation/package_qa.json",
+      "bytes": 608,
+      "sha256": "1a7228567fa1fe0af2f4255e97deda8dce6da705b07b4af6d509f282cb2c968d"
+    },
+    {
+      "path": "evaluation/runtime/cpu.json",
+      "bytes": 10730,
+      "sha256": "9159d0d8c0be3e318625468b77965c2360f86a2236524edd2aedabc0f837276a"
+    },
+    {
+      "path": "evaluation/runtime/cuda.json",
+      "bytes": 10894,
+      "sha256": "655913e46e21520550f19e3e19b6b280b2757a1a37dc0f305e8c93575094e43f"
+    },
+    {
+      "path": "evaluation/signal.json",
+      "bytes": 213439,
+      "sha256": "f88eebb932da1e15dc646d6545f4739949144ea09bc961cd02c30f9a40007a50"
+    },
+    {
+      "path": "evaluation/speaker_consistency.json",
+      "bytes": 12239,
+      "sha256": "e8df4cf787396b756721ba7672b384046ec837014e9b9298e04e1a327ac0a620"
+    },
+    {
+      "path": "evaluation/utmos22.json",
+      "bytes": 171875,
+      "sha256": "bb1f78fd36bbcf928aa71ca7376ff6d948f9e7e613c424caf5ce84e9e0ecf20d"
+    },
     {
       "path": "inference.py",
       "bytes": 6252,
@@ -32,14 +97,19 @@
     },
     {
       "path": "inflect_nano_v2_frontend.py",
-      "bytes": 10246,
-      "sha256": "9f800c7d86939b6eadbdec723d2411caaab820035f87de72ca8bece1a0ab3312"
+      "bytes": 11687,
+      "sha256": "530825406a4064ff8c7a873b35eb77920823734a98e77180e2aced48a6868dca"
     },
     {
       "path": "inflect_vits_frontend.py",
       "bytes": 2031,
       "sha256": "70eb4cdb96b2752015f72be984f5e62e0f26d7dc80e708875601702057b95cc2"
     },
+    {
+      "path": "LICENSE",
+      "bytes": 10455,
+      "sha256": "12c48c640d1d0baf158d45175a4238b9227c71c77f50b0d15791b8efc08cd793"
+    },
     {
       "path": "model.pth",
       "bytes": 37529995,
@@ -50,6 +120,11 @@
       "bytes": 443,
       "sha256": "4c3d2d7c361792e3fe398745989dbf166dde40193122afdfbe8d2b1538b307ae"
     },
+    {
+      "path": "README.md",
+      "bytes": 7126,
+      "sha256": "bfa9b01e2a571a83d2d59510ced6cd38c4e045255a80ed5fcf6fdb9b9a977ca3"
+    },
     {
       "path": "requirements.txt",
       "bytes": 128,
@@ -85,11 +160,6 @@
       "bytes": 206,
       "sha256": "16ca4890f3e0107250b4a522c0127779994a5a2b99fc35d6a27f93496a72153a"
     },
-    {
-      "path": "runtime/text/LICENSE",
-      "bytes": 1053,
-      "sha256": "f89accd9d66f6a011f19997d7a21cf71a35ffe2e2d0911b5460cbd0cab52a581"
-    },
     {
       "path": "runtime/text/__init__.py",
       "bytes": 1572,
@@ -100,6 +170,11 @@
       "bytes": 2880,
       "sha256": "11dc7325547290e529c597e620f3aaf8d0bd8a2a2987d48eb42d2289534e42d1"
     },
+    {
+      "path": "runtime/text/LICENSE",
+      "bytes": 1053,
+      "sha256": "f89accd9d66f6a011f19997d7a21cf71a35ffe2e2d0911b5460cbd0cab52a581"
+    },
     {
       "path": "runtime/text/symbols.py",
       "bytes": 642,
@@ -115,6 +190,31 @@
       "bytes": 7526,
       "sha256": "db11ba9b5ffd342abffc59229de7575a7d9e23468f91c6398e70eb4e01636716"
     },
+    {
+      "path": "samples/male/conversational.wav",
+      "bytes": 180780,
+      "sha256": "f9e0f1d0ff2c5ced4ac05d8bc44b8a14a8b0c79ea36ed5aaa936df1840e899ed"
+    },
+    {
+      "path": "samples/male/names_places.wav",
+      "bytes": 182828,
+      "sha256": "fdeea745130c5d873fcb6ea8a5674d529c325739b2ad307fa422e7e2c9fd4997"
+    },
+    {
+      "path": "samples/male/numbers.wav",
+      "bytes": 229932,
+      "sha256": "a73cc26c31c11b1a78d26f8ebd382de4efb37b3514d7f2c91b9528ca79e7daf1"
+    },
+    {
+      "path": "samples/male/punctuation.wav",
+      "bytes": 255020,
+      "sha256": "ca433246649b7588982afa6dd67db2f40edf238b8ef87f865d8bb3eb885f386a"
+    },
+    {
+      "path": "samples/male/technical.wav",
+      "bytes": 192556,
+      "sha256": "2b59ad3d449c22b851d4a5f882f7a5e9d3413d0bc1e1037aa3374a3470c91114"
+    },
     {
       "path": "smoke/manifest.jsonl",
       "bytes": 275,
@@ -144,6 +244,11 @@
       "path": "third_party/VITS_LICENSE.txt",
       "bytes": 1090,
       "sha256": "1cdf858b4d555567398b378b4a0166e97f7908c56607101b8bf9bb6754cdff3f"
+    },
+    {
+      "path": "THIRD_PARTY_NOTICES.md",
+      "bytes": 892,
+      "sha256": "ef5343cd39713d313aefcfd09bb55fd6606ae2dacabeee13e0abdaa2e5b62589"
     }
   ]
 }