TTS text normalization speaks "sil" in US-style dates ("May 3, 1999")
Los mantenedores suelen responder en 1 día
Nadie ha tomado este issue todavía.
Evaluación
- Dificultad
- 4/5
- Tiempo estimado
- 3-5 días
- Aptitud para principiantes
- 38/100
Línea de trabajo
The payload names no source files. Start by reproducing with the bash script in the issue, using nemo-speech synthesize with --tn-model-dir on the "May 3, 1999" input, then transcribing the output. Find where the text normalization date grammar's pause token reaches the tokenizer as a word, since the issue's hunch is that it leaks. Done when the "May 3, 1999" and "Oct. 8, 2026" cases transcribe with no "sil" or "SIL" and the other formats still read correctly.
Escrito por el modelo de indexación a partir del texto del issue.
Descripción
With --tn-model-dir, MagpieTTS says an audible "sil" between the day and the year of a date written as "Month D, YYYY". Without the comma ("May 3 1999") or day-first ("3 May 1999") the date is read cleanly, and so are the same words typed out ("October eighth, twenty twenty six").
Impact. This date format is everywhere in English text, and normalization is what makes MagpieTTS read numbers at all, so any app reading text aloud has to choose between this artifact and dropped digits.
Expected: "May third, nineteen ninety nine", with at most a pause at the comma.
Actual: "May third sil nineteen ninety nine".
Reproduce. NeMo-Speech.cpp 0.2.0 on macOS arm64 (Metal), tn_configs.tar.bz2 from the v0.2.0 release, MagpieTTS 357M, voice John. Needs only nemo-speech and curl; the transcripts come from Nemotron 3.5.
#!/usr/bin/env bash
# Reproduces: with --tn-model-dir, MagpieTTS speaks a "sil" between the day and the
# year of a US-style date ("May 3, 1999"). Needs nemo-speech 0.2.0 on PATH, curl, and
# the magpie and nemotron-3.5 models (nemo-speech pull magpie; nemo-speech pull nemotron-3.5).
set -euo pipefail
VERSION="$(nemo-speech --version | awk '{print $2; exit}')"
WORK="$(mktemp -d)"
trap 'rm -rf "$WORK"' EXIT
curl -fsSL "https://github.com/NVIDIA/NeMo-Speech.cpp/releases/download/v$VERSION/tn_configs.tar.bz2" |
tar -xjf - -C "$WORK"
say_and_hear() {
nemo-speech synthesize "$1" --tn-model-dir "$WORK/tn_configs" --voice John --seed 6285 \
--quiet --force -o "$WORK/out.wav" >/dev/null 2>&1
printf '%-34s -> %s\n' "$1" "$(nemo-speech transcribe "$WORK/out.wav" --quiet 2>/dev/null)"
}
echo "nemo-speech $VERSION, tn_configs from the same release"
say_and_hear "It happened on May 3, 1999."
say_and_hear "The launch is on Oct. 8, 2026."
say_and_hear "It happened on May 3 1999."
say_and_hear "It happened on 3 May 1999."
nemo-speech 0.2.0, tn_configs from the same release
It happened on May 3, 1999. -> It happened on May third **sil** nineteen ninety nine.
The launch is on Oct. 8, 2026. -> The launch is on October eight **SIL** twenty twenty six.
It happened on May 3 1999. -> It happened on May third, nineteen ninety nine.
It happened on 3 May 1999. -> It happened on the third of May nineteen ninety nine.
Our (Claude + me :D) hunch, unconfirmed: a pause token from the date grammar reaches the tokenizer as a word.
- Lenguaje dominante
- C++
- Estrellas
- 181
- Forks
- 42
- Merge medio
- 1 d 8 h
- PR fusionados (30 d)
- 8
Preparar el entorno
- Sin Dockerfile ni archivo de Docker Compose
- Sin plantilla de pull request
- Leer la guía de contribución
Primeros pasos
- Lee el issue completo y luego la guía de contribución del proyecto.
- Comenta en el issue que vas a ocuparte — evita que dos personas hagan lo mismo.
- Haz un fork del repositorio y trabaja en una rama.
- Abre un pull request que haga referencia al número del issue.
Más de NVIDIA/NeMo-Speech.cpp
-
Dificultad 4/5 3-5 días Aptitud para principiantes 20/100
NVIDIA/NeMo-Speech.cpp#67 ·
Los mantenedores suelen responder en 1 día
-
Dificultad 3/5 Medio día Aptitud para principiantes 66/100
NVIDIA/NeMo-Speech.cpp#64 ·
Los mantenedores suelen responder en 1 día
-
Dificultad 4/5 3-5 días Aptitud para principiantes 52/100
NVIDIA/NeMo-Speech.cpp#61 ·
Los mantenedores suelen responder en 1 día
-
enhancement
NVIDIA/NeMo-Speech.cpp#56 · 3 comentarios ·
Los mantenedores suelen responder en 1 día
-
Expose confidence estimation / token probabilities (currently always 1.0)Posiblemente ocupada @pskrunner14 la tomó hace 15 días. Abiertoenhancement
Dificultad 4/5 3-5 días Aptitud para principiantes 48/100
NVIDIA/NeMo-Speech.cpp#55 · 2 comentarios · 1 asignado ·
Los mantenedores suelen responder en 1 día
Todos los issues de NVIDIA/NeMo-Speech.cpp
Issues similares
-
Dificultad 2/5 1-3 horas Aptitud para principiantes 70/100
FriendsofECCE/ECCE#336 · 1 comentario ·
-
Dificultad 2/5 1-3 horas Aptitud para principiantes 66/100
bitcoin-core/gui-qml#1030 ·
Los mantenedores suelen responder en 3 días
-
Dificultad 2/5 1-3 horas Aptitud para principiantes 82/100
Los mantenedores suelen responder en 1 día
-
Dificultad 2/5 1-3 horas Aptitud para principiantes 85/100
tesseract-robotics/tesseract#1384 ·
-
Dificultad 2/5 1-3 horas Aptitud para principiantes 62/100
NVIDIA/nccl-extensions#26 ·